What happens when a PhD Professor in Analytics launches an AI Writing Product?

4 avril 2023 · AI Product Creators

00:17:43

Martin Pichlmair is the CEO of Write with LAIKA, Associate Professor at ITU Copenhagen and Co-founder of Broken Rules. He Holds an PhD degree (Department of Informatics) in Vienna University of Technology. In today’s episode, Martin explains that LAIKA is designed to make AI-generated writing more accessible and user-friendly, with the AI and the user working in a tight interactive loop. Martin highlights that their product uses a "no-prompt" system, which means users don't need to be skilled in prompt engineering to get meaningful results from the AI. Instead, the software handles most of the prompt engineering behind the scenes, making it easier for users to interact with the AI. Tune in to hear Martin's insights and experiences in building LAIKA and how you can apply these lessons to your own product.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Martin Pichlmair:

• LinkedIn: https://www.linkedin.com/in/martinpi/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
welcome to the call, Martin. Thank you for joining. Tell us a little bit about your product.

Martin Pichlmair:
Okay, so I'm Martin. I'm the CEO of Write with LAIKA. And our product is a kind of creative writing tool that is using large language models, in our case, quite small, large language models to, support writers when they get stuck or when they need more text, that is influenced by their previous writing.

Dhaval:
Wow. Okay. So when the writers get stuck or when they. Interested in continuing with the style and the tone of their previous work. They can use your product including the contents, the storyline, or anything along those lines.

Martin Pichlmair:
Yes. How LAIKA works is that you you upload existing writing. You have when you get, for example, stuck in a murder mystery because you don't know who the murder is. Funnily, we had that case twice already with users. And then you upload what you have written before and our, system fine tunes a language model With your text and then you can prompt the model to continue writing in your voice, in your using your characters. You mentioned using scenes you have been writing about in the past and very much sounding like you. Now you can do that with your own text. Or with the text of famous writers, we have, for example Dostoevsky in there and Jane Austen in there. And a lot of, all of them, of course, dead and out of copyright writers that you can also collaborate with in a similar way by asking them how they would continue a sentence, for example.

Dhaval:
Wow. So it has memory and context as well as style and the personalization built into it. So is that. Large language model that's very different from Chat GPT 3, which would spit out very confident phrases very long phrases. But they're also having the same style. Is that, how is that different from the large language models? You said that you have used large language models or you have you built on top of them or like, help us a little bit on how have you built this.

Martin Pichlmair:
So we've built this on very small, large language models. They're still in the same architecture and come from the same family, but they're very small because that gives us the ability to fine tune them very quickly. It takes like five minutes. If you upload , a half done book, for example, takes five minutes and you get your own, we call them brains because that's a nice metaphor. Your own brain based on your writing to interact with. Now, of course it has an understanding of the context, but it's not always super, like it doesn't have an actual understanding. It can just play with probabilities of words, just like all of those language models do.

Dhaval:
Wow. Very cool. Let's dive a little bit into your product journey, is this your first startup? Is this your first AI product? Tell us a little bit about your background, Martin.

Martin Pichlmair:
So I have a weird background. I did a PhD in computer science originally at the University of Vienna, at the tech university, and then worked in academia for a couple of years. I got a little bit, I don't know I wouldn't say bored, but I wanted to do something differently. So I started a video game company and then after a year started another video game company because the first one didn't work out. It didn't work out, but it also didn't not work out. It was fine. It was just not meant to be a longer existing thing. The second one actually is still around. It's called Broken Rules and makes awesome in the games. But I'm not involved anymore because I decided at some point to go back into academia. So that's where I spent the last seven years until last year where I just realized with my partner, That we have a huge connection between what I was doing in research, which was using generative AI to create systems for video games and her background, which is writing for video games. So we sat down and, uh, started workshops during the Covid Pandemic when everyone was sitting at home. We started online workshops where we introduced writers. To, the newest possibilities in language models, using very, very clunky tools at that point. And after three or so of those workshops, we realized the workshops are always poked out, but it's really hard to work with the tools that are there. So we decided we have to make our own tool, and that became a research project that was funded by the Danish state, uh, in the beginning. With the intention of turning it into a product. And since last November, we founded a company and turned it into an actual university spin off that is based on yeah, research that is now working on a product that we will commercialize within the next month

Dhaval:
Very interesting background. You do have like a very traditional computer science background, making you very competent in this area. Right. So, quick question. You mentioned that you launched this in November, but you haven't commercialized. It doesn't mean that the product has not launched yet.

Martin Pichlmair:
Yeah. We have a wait list and we have, a data with, nearly 2000 users. So there are a lot of people using it every day, but, it's not a commercially launched yet. We're still only free for select users.

Dhaval:
Very cool. Is this. Is this a self-funded or have you bootstrapped this whole thing? Are you intending to, or have you raised capital? And are you intending to raise capital as you move forward?

Martin Pichlmair:
Well, we got some funding from Danish State again. The program that we're in that funded turning research into a product last year that was still in the context of my university has a follow up program that funds your salary basically. So we are kind of weirdly half bootstrapped. We have no investor, but we have, our salaries covered by the Danish State but a very low salary. But still, it's good enough to, know that we'll be, we'll be around for another year at least while this funding runs. we're looking for investment in the moment. It's, we are talking to a lot of VCs. It is. It just takes a while it seems.

Dhaval:
Yeah. how is that playing out in this current market? Like how is that, can you give us have you done this before? And if you have, like how is it compared to the current market, if you can speak to that.

Martin Pichlmair:
So I haven't done it before, but a good friend of mine has a very similar company, actually a very different company, but also an AI company also in Denmark. And he also has an academic background. It's otherwise very, very different because it's B2B and started out much bigger than we are. But it looks like they, the climate they saw two years ago is very different to what we have now. So I'm getting all my tips from him and half of them don't work anymore. The climate is not good in the moment. Even in the hype space of creative ai, there is a lot of chicken egg problem, happening in the sense that investors wanna see. They actually want pay traction very often .They want to see some pay traction or immense numbers in weightless users or something like they wanna have proof of actual viability very early on. But it's such a new area that you are actually creating a market. So it's very hard to say where this whole journey is going because the whole, like AI is not super new. But generative AI is really something that is only a thing since like a year or so. It's very hard to say where the journey goes in the moment and, like it could all still just be overhyped. Then I would understand the need for having paid traction, but it could also be that we are just opening, creating a completely new market here, and then a little bit of trust would be nicer than having to prove things too early.

Dhaval:
Yeah. Yeah. Where are you in the product stage in terms of product development? Are you close to? I know you mentioned you were ready to launch in a few months. Are you close to finishing your product development? Is is that almost there? Like, if you can share that. Cause my follow on question's gonna be on, what were your top learning lessons around creating an AI generated product? What was that experience? What was the top learning lesson there?

Martin Pichlmair:
So I think the whole idea of finishing a product is not really how it works with software as a service anyway, but especially in this extremely fast moving area of, AI in general, but especially generative AI where new technologies come out on a sometimes weekly basis. There's a lot of competition, but there is also just a lot of speed of development. I don't think we will ever be at the moment where we say, now we are done with this. Instead, what we are, where we are trying to get is to a point where we can say, this is our 1.0 version and we hope to be there in the month actually. And, from then on we of course continue building. and one of the main challenges to my surprise actually, and maybe that was the main learnings, is since what we're working with is so new, it is really, really very hard and crucial to explain to users what we even doing here, because they are very curious, but they have a very hard time understanding these new ways of interacting with the computer. And, it takes. Nearly as much time to make the product as it takes to actually package whatever you're building in a way so that an everyday user can understand what is even going on here. So that's quite challenging and that is something that we also are constantly in the process of figuring out.

Dhaval:
Yeah. One of the biggest challenges for generative AI products is gonna be around user experience and managing that human AI interface. What has been your biggest learning lesson in that area for your product?

Martin Pichlmair:
So we have a nearly no prompt system. So most, in most AI systems, how you interact with what a generative AI systems. If you interact with them, you do some kind of prompt engineering as they call it, so you get better and better at formulating what you wish to get out of the model. More and more precise over time and you just, it's a learning process. You prompt them with text usually, and you just get better at describing what you want in the way the machine understands. which is an interesting interaction, but it's actually surprisingly technical. And our approach is a little bit, since we are all about making this more accessible, these systems is our approach hides that. We are, of course also prompting in the background very much from the user and tries to make sure that you get a higher percentage of. Good results automatically instead of giving you this huge space of freedom that you have with more technical systems. But those systems exist so if someone is technically inclined, they can just use a different system. It's just not what we are making a software for.

Dhaval:
Yeah. You mentioned something about prompt engineering and how does that play out when users are engaging with the product, do they get to like decide exactly what's gonna be the better prompt to ask, or is that something done behind the scenes?

Martin Pichlmair:
Most of it is done behind the scenes. So the base use case of LAIKA is, you upload your materials to train with your voice, and then you just write half a sentence or a full sentence and a question mark, for example, and then you ask the system how it will continue from there. Now, of course, technically you are just prompting. But in practice it feels like you are writing and now, and then you give a little bit of control to the machine and then you override the result immediately and work in this very tight interactive loop with the system. So it's a bit of a, yeah, more. There's a lot of discussion about generative AI in the moment among artists that, uh, somehow do not appreciate the fact that there are so many machine generated images out there and what we are doing is very different because there is no result. That is trust. What comes out of the generative ai, it is always the author with. Their own thoughts remixed by the machine creating something. So it's very much 80% is human authored and what is even made by the machine vanishes, hopefully. So it's a slightly different way of interacting. It's using the same base technology and the same post model base modalities of how you interact with a machine learning model. But it hides them and it makes the loop even tighter than existing systems.

Dhaval:
That's where the magic is, making that loop tighter. Thank you for sharing all your knowledge. Martin. Anything else that you wanna share about the future of your product? Where do you think it's going? Do you have a big vision around where it's gonna be in about 10 years from now?

Martin Pichlmair:
About 10 years. Oh boy. Well, in the end I think that I actually have a big vision how, where it will be in 10 years or where society will be in 10 years. I wrote a blog post about it recently. That's why I have. The thing is I actually think that a lot of interacting, interaction with generative models is not so much about getting results and much more about the experience. So what we are doing is we are interacting with cultural, the cultural heritage of humanity, in a live and, uh, uh, back and forth ways. So we synthesize something that uses images that are up to thousands of years old, like the photos of those images, but still human expression that is hundreds, thousands, and so on, years old, and we synthesize them into something new. And that activity is actually a very interesting activity of interacting with the, what humanity is with all of our culture. And I think it will become just something very normal for us to do. That we're not only interacting with the way a specific historian has recorded our history, but in a much closer, again a tighter loop, interact with past images, past writing, past words, past figures, maybe historical figures and so on in a, and of course, synthesized way, but still very much informed by who humanity was in the past. And I think that just will lead to a different understanding of who we are. And that will maybe take 10 years, but it's gonna be interesting. That is a really big vision. I know.

Dhaval:
Yeah. That's amazing. I, I look forward to get to there as well Martin Write with LAIKA 2000 beta testers, how many people who have signed up for the wait list.

Martin Pichlmair: Well, not many more.

Dhaval:
Not many more. Okay. And you are ready to launch your product in about a month's timeframe and, you have exciting, future ahead. I am looking forward to following your journey. Martin any last thoughts on anyone who wants to start on this journey? AI, product creator who wants to build an AI product. In generative AI space using text as their, as their tool. What would the, your advice be for them? any tools that you would recommend for them to use to get started?

Martin Pichlmair:
I think the important thing is to find a niche, because I think in the very near future, like next year, most likely, Microsoft will trust, integrate generic functions for creating text into all of their products. And then that's just gonna cover 90% of the use cases. So if you want to get started in text, it's like you have to find a very, very, very specific niche, maybe in the business to business sector, to be able to still be around in a few years when that will just be a standard function of word processors. So I think that is my main tip to just. Yeah, keep it, keep it focused.

Dhaval:
Now in terms of execution to keep it focused, would that be custom models? Would that be built in tune with GPT 3 or GPT 4, whatever is the next version? How would you suggest that?

Martin Pichlmair:
I do like our autonomy. We own nearly all of our stack, all of our core stack is owned by us. And that gives us the ability to define our own policies of how people can interact with our system it gives us safety. It makes us independent of the server costs of other people. Of course we are hosting it in the cloud. It is a little bit cheaper also in the moment. So this autonomy is something that works very well for us. Now, I think it's of course, expensive to build a product, using your own stack, but I think there is will just be more and more need for diversification of what you're doing. And we can have more diversification than any product that builds on the vanilla GPT 4 version or 3 or 3.5 or whatever.

Dhaval:
You cannot really have that niche if it's built on top of a vanilla version. Well, thank you so much, Martin. It's been a pleasure having you on the show. I'm looking forward to following your journey,Thank you so much, Martin. Have a great rest of your day.

Écouter Écoute encore Continuer Écoutez...
S’abonner Se désabonner
Partager

Episodes

Transforming Creativity and Content Creation with Zach Hanson - A Deep Dive into the Future of AI Product Management
19 mai 2023· AI Product Creators
Zach Hanson is an expert in artificial intelligence and machine learning product management, with experience developing AI solutions for Fortune 500 companies including IBM, Brightcove, Capital One, and Wells Fargo. He holds degrees from the College of Charleston and Johns Hopkins University. In today’s episode, We discussed power of AI, Zach discusses how it aids in tasks like content parsing, summarizing, and producing video trailers. He also explores the interconnection of different AI models, and the rise of content generation through freeform speech. We discusses how AI technologies, like ChatGPT and GitHub co-pilot, can streamline creative content creation and refine stories or code. Finally, for those looking to enter the AI product management or creation space, Zach advises building something to understand core product fundamentals and getting comfortable with data. Tune in to hear Zach Hanson's insights and experiences in building Inworld AI and how you can apply these lessons to your own product.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Zach Hanson:

• LinkedIn: https://www.linkedin.com/in/zachary-hanson-a1a761a3/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
Hey, Zach, tell us what you've been up to.

Zach:
So everybody, I'm Zach Hanson. Dhaval it's great to see you again. You and I used to work together in a, past life in the AI field but What I've been up to lately is pushing for AI innovation within video, specifically at Brightcove, which is a great company, and working around how we actually build out and better experiences for our customers in the video space.

Dhaval:
Oh, wow. That's like the most cutting edge space in AI right now. The video and AI generated videos and all of that as we speak in April of 2023. What specifically is Brightcove's mission? And, yeah, if you could share a little bit about the, what is the product, what pain is it solving, and what are your customers, who are your customers? And then a little bit about where you are in the product space, like in the journey. Like, are you a startup? Are you, have you already found the product market fit? Are you enterprise? So yeah, any of that? So, a lot of questions, but just trying to understand the product story.

Zach:
So it is interesting, right? So we're a very unique company. So Brightcove's goal, to answer your first question, has become the most trusted company in streaming. So that's our goal, right? That's the big headline. That's what we're pushing and we kind of run the line of being a media company and an enterprise company. So from a competitive landscape perspective just to frame where we're at, would be the Vimeos of the world, Kaltura and some other companies that are providing OTT services to brands around the world. And Brightcove has actually been in the game for over 15 years. So to answer another piece of your question is we're a publicly traded company. We've been around for over 15 years and we've been providing these services for live streaming, for video on demand OTT for years, and it's pretty amazing. So we're one of the biggest little companies you might not have heard of. But with that comes a lot of responsibility around data because we actually ingest around one to two petabytes of video data every month. So we have an absolutely enormous catalog and data warehouse of videos, audio, all sorts of content that is being leveraged by our customers. Now, to answer one of the other questions rolled in there about our customers and some of the ones we can talk about, we help deliver video, great quality video with our encoders for the Olympics the year ago. Wow. Yeah, we've done that. We work with south by Southwest. So folks, if you've watched conference video from there, that's Brightcove under the hood masterclass for instance. So we have a huge list of really amazing clients who are doing All sorts of different things with video. And that's what we're trying to enable. Now, the other piece of your question, where are we at in the product journey? Are we a startup? Are we a mature company? And I would say from the delivery of video, we're very much a mature company. But when we start to think about machine learning and leveraging Models that are out there building our own models, we're really much more in that startup phase where we're trying to find the appropriate product market fit for the different types of models we might build or leverage the really immense amount of data we have to train models and do some really cool things.

Dhaval:
Wow. Thank you. Thank you so much for answering all those questions. I threw a barrage of questions at you. Wanna dive in a little bit on your last answer here on the topic of being new to AI ml. And what I wanna understand, Zach, is what is the customer trying to do when they want to use the AI ML capabilities for Brightcove? What is the thing that's going on in their head when they are like trying to use your specifically AI ML features?

Zach:
So this is where it's also like an interesting story because there's a lot of stuff that we're focused on from a machine learning perspective, kind of under the hood, things that our customers might not know is being powered by machine learning. So some of that has to do with encoding and how we get the video to the actual end user in a very efficient manner or in an efficient manner. As far as doing CDN optimization and making sure that the ultimate end user, which is our customer's customer, whoever's watching video. Has a great experience and that's where the bulk of our effort's been. But when you think about pain points, as we think about becoming more of a media company, when we think about enabling producers of content to be able to do some really cool things there's really this kind of crawl, walk, run approach. One is when somebody uploads content to our catalog or their catalog through Brightcove, there are sorts of metadata that should be tagged in those videos. Oftentimes people are having to do that manually time stamping stuff or putting this as a certain piece of a sub catalog within their overall experience. So we're trying to do some automation through their of automating tag management to suggest to our customers tags they might need in order to ease the burden of some of the metadata management, but then you go up the chain to content itself, the video, and we start to think about object recognition and video. We start to think about segmenting video. So you can easily cut and pull out specific elements of a video. For instance, if you were watching a soccer match or football match I grew up in the United States, so I'm a little bit more used to American football and I've become a bit more of a fan of the universal football in the years, in the past few years. But you might have an hour and a half long game and only have two goals or none. So the ability to be able to search through a video and find that really intense moment where somebody actually scores a goal, be able to rip that out really quickly and repurpose that content for marketing is very powerful. And there's a lot of startups actually playing in that space and then you have the bigger players like ourselves, Brightcove. Then you also trying to play around with segmentation of video. So it really runs the whole gamut where we've been focused mostly on backend support, leveraging ml. All the way to that kind of front facing customer content production type of use case.

Dhaval:
Yeah, I, that's amazing. You have, I can think of so many use cases. I was at a photo shoot video shoot event this weekend where I was hosting it, and we have like terabytes of video content that we created and now I want to create recap videos for that event. And I can imagine being able to feed something like that to your platform. Is that, am I getting it right? Would that be a potential use case? Is that how you It is parts out valuable clips.

Zach:
Exactly. And that is part of a potential use case that we're exploring. But this is where it goes back to being in that kind of pseudo startup space. Like with all the data we have. With the great customers that we have, there's a lot of opportunity there and we're still in that feeling out phase of saying, what are those pain points to your other question. And like the use case you just gave, that might be something at the top of mind for a lot of our customers. And that's where we're just starting to put the feelers out and understand how we might be able to build some of these things out and make sure we have the right product market fit before rolling something out to our broader customer base.

Dhaval:
Yeah, that's very interesting. Just like thinking for like content creators like myself. That event is an example of a use case. This podcast is an example of a use case, parsing out insights from this video, insights, and then publishing them. And then for courses that I create on product management and artificial intelligence, it's the same thing. I can imagine being able to give you a whole course and create a trailer for that. So there are a million use cases that you can be going after. Are you thinking of like any big use cases right now that you are willing to share with us, that you may wanna pursue in near future?

Zach:
You know, none that I want to talk about specifically for Brightcove. But there are a lot of really interesting things in that segmentation space that just interest me and that might be wrapped into something we end up doing with Brightcove. It might not, but meta just came out with a paper on Segment. Anything. Have you heard of this model that they've built? Again, it's in that segmentation space on video or, things like that. But with all the other models, stable diffusion. People are starting to piecemeal these different models together to come up with really cool use cases of really just actual content generation where you could take freeform speech to say what it is that you want to put together. Like for instance, it could be a thumbnail. Even a video thumbnail for the conversation we're having now and with stable diffusion, with Segment, anything you might be able to pull an appropriate clip, you might be able to change it to meet and give a visual for this conversation that you know, you would normally have to go and pay an artist or get a graphic designer to pull together in the snap of a finger. So there's a lot of Git repositories out there that's like edit anything which pieces all these models together and they're just a lot of really interesting work being done there in this space. That makes me excited and makes me excited to see how we might be able to leverage these different technologies just in video in general.

Dhaval:
Yeah. You are in a hot space, like what you just said is like so much happening there and being able to. Creators are living in such a good time right now. Like they can have great idea, great creative ideas, and the execution is streamlined to the max. Almost to the max, right? So as we go forward with this progression of technology in this space, Zach, Do you envision a future where creativity is a lot more valuable thing? Or do you see that being, co-piloted with ai. What is the value of creativity as we progress with this technology that is trying to commoditize it at the same time?

Zach:
Man, that is a philosophical question that is hard to answer now. I consider myself a creative, like you can see behind me, I like to write books. So, from that perspective, I like to flex the creative part of my own brain with coming up with stories for fiction books or writing non-fiction books. And as ChatGPT has come around, I of course have kind of played like giving certain prompts to see if it would give a similar storyline as to something that I've come up with. And of course, When you prompt it in the right way, it can get pretty close, but there's always a little bit of a gap. I think there is a premium on human creativity that, I'm not as bullish on thinking that it's going to disappear with the advent of a lot of these great technologies. I think there will always be that value for that human perspective. But like with GitHub co-pilot, with ChatGPT, I think what is happening and what will happen is the. Ramp up or the curve to get creative content out there will probably be reduced. Meaning if you can leverage it in an appropriate way to bust through rider's block to Refine a story a little bit more quickly than you would if it was just pen and paper. Same with code. If you can leverage co-pilot in a way that allows you to get 80% there and then really buckle down to make sure you have the finer points really well fleshed out, I think it's amazing. So I think it's going to actually more unlock the ability for people to be creative. A lot more at least more during the day. So I think it's actually a great tool. So I'm optimistic, I'm not a pessimist.

Dhaval:
That's great, man. I think in the same way, I think as this tools advance, we will be seeing, a breakthrough in creativity for people who don't think they're creative. It's gonna create that lane for people they're not necessarily seeing themselves as creators or creative people. It will enable and empower them to be creative and we'll see a whole breakthrough in creative potential of humanity as this unfolds. I'm looking forward to that future myself.

Zach:
And, on top of that, so similar to you, like I teach a course in data product management at Boise State University yesterday in my class, which is all senior level software engineers, computer science students. A lot of them are working right now and just doing school part-time. And one of my students talked about chatGPT, and he had a non-technical product manager on his team who ended up using copilot to just try to explore a little bit more and understand what his team was doing. And he said it was actually amazing, was co-pilot. He was able to get 80% there. Now the product manager lacked a few of the key core computer science fundamentals to get there. But it was just a little bit of a push from my student and his co-worker to get them all the way. And he was like, it was amazing. He's like, now our product manager has a better understanding of what it is we're doing. So I think it is also another opportunity for product managers and as product HQ to be able to. Empathize with their developers, more empathize with machine learning engineers because they can go out and explore on their own with a little bit more ease. And I think that's an amazing potentiality for growth in the product space.

Dhaval:
Yeah, I love what you just shared. Thank you so much. Changing the gears a little bit, what are some of the product market fit lessons you are learning as Brightcove? Or any of your other advantages is starting to integrate AI into your product.

Zach:
Yeah. I have a lot of these just from years past. Right. Overall, A lot of companies and you and I have been in this game for a decade, right? So it's now becoming I think easier to integrate ML into products, expose that to customers and have them appreciate it. But, my experience has been in the past five, 10 years is that oftentimes. Coming in as a machine learning product manager, you tend to push for things that are machine learning related, when oftentimes a simple set of rules might do the exact same thing and be more cost effective. So I think in past roles I've pushed for ML when really it might not be needed, or we might be a little bit ahead of the curve. So I think there's even today something to be said about. Hitting the pause button as you're thinking about implementing ML into any product you have. Like, is that the right solution? And oftentimes now it might be, but taking a step back and saying, can we do this more simply in a more cost effective way, that's not gonna take up more compute power. It's not going to need to have a whole model development and deployment pipeline tied to it. So, thinking about it from that perspective and making sure the market is actually looking for an ML solution is key. Because I've been hung up on that several times in the past where we overbuilt for something that was actually quite simple.

Dhaval:
Yeah. And like you said, now it may make sense to. Do ML or AI because of the ease of doing it. Right even then though, like, I like what you just said, which is still take a step back and see if you really need it. Are you just getting sucked into the hype or does your customer really benefit from you doing it? And that's an important distinction, like that's a really important distinction you are making there. Thank you for that. What are some of the, specific challenges that you are running into in terms of working at this large organization that is trying to find this new foothold with this highly innovative use cases? You may reframe them as opportunities rather than challenges. But yeah, however you can, share that. I would love to hear that.

Zach:
Yeah. The answer is data, data, data, data. And you know, it is both opportunity and it's both a little bit of a road blocker. companies that have been born into the cloud era that have been born with data first principles you can see are really getting ahead right in the ML space because they have. Well kept data with great metadata tied to it. They have strong taxonomies. It's well annotated so you can build models relatively easy or build features to put into models and it's great. One of the things that I've struggled with across the board and a lot of the companies and I've been a big financial institutions even with Brightcove, that's a 15 year old company now. Is making sure that we're treating data as a product. That's a big thing for me. It takes part of the data mesh principles, which is if you are building a product and deploying it within a company, it could be Brightcove, it could be IBM it could be anywhere. Oftentimes, that product that you're gonna be developing is also gonna have data as an output. And that cannot be an afterthought because what happens is things start to metastasize and you get bigger and you move on to the next product, and then that produces data. And if you don't have technical product managers and data engineers really evaluating the outputs of those data. And making sure that it has a strong taxonomy, making sure that the metadata tied to it is clear, concise, usable, traceable. Then you're gonna be at a disadvantage down the road when you actually want to implement machine learning practices and machine learning models. Because you're gonna spend a lot of your time going back and trying to clean that data or introducing a third party vendor to come in and either clean it, annotate it, and get it in a position where you and your machine learning engineers can take advantage of it. So for me, it's all about data, and that is an opportunity for everybody, is to sit back, look at the data that your product's already produced, and what you have available to build off of, and making sure that it is as clean as you can possibly have it. Implementing governance across the board to make sure your teams are treating that data well. And that way you've heard the, the trope about now people have data lakes and then often turning into data swamps because of that lack of governance. You wanna avoid that. And that's the biggest opportunity that we have. And the biggest opportunity, I think, really any company across the board could have if you want to take advantage of machine learning practices as you grow.

Dhaval:
Yeah, data is still as relevant as it was 10 years ago and it's never going to be less relevant no matter how advanced AI gets. So one follow up question is slightly, slightly different question rather, is, what are some of the opportunities for. Brand new people like trying to get into this space who don't have prior AI ML background. I know you have been working in the space for 10 years. You've been doing this for, before. It was cool before you and anyone knew that there was something called AI product management. You were an AI product manager. Right? But for people who are just like oh this is not going away. I'm stuck with this field forever. Now I better learn it. It's like it's the new software. Right? So what are your. Recommendations for folks like that who want to learn about getting into AI product management or AI product creation space?

Zach:
Yeah, first and foremost, I got into AI product management by happenstance. You know, for me it started out as being an entrepreneur. So I still think the greatest base for getting into product management, whether that's a data product manager, an AI product manager, is go build something. Get down core product fundamentals, which is for me, design thinking, understanding how to go from a broad problem space, refining that in the personas that are affected by it, understanding what they're doing today, what you might be able to change to make their lives better, and then executing on a product. So that's was my journey, was building a product that ultimately failed. Before I went into industry and then found my way into the machine learning space. But number one, build something. It doesn't have to be a unicorn. It can be something very small that either sell or give to friends, family, whatever. Build something on top of that. If you are keen on making sure that you are gonna be doing ai product management is get comfortable with data. Again, we just talked about data being the core of everything. Yamak Dajani has a great book out around data mesh that covers a lot of the principles that roll over into machine learning and ai. It's making sure you have federated governance around your data. It's treating data as a product, so applying product fundamentals to the data you might use in machine learning. It has all the other elements that you might need to be a good AI product manager with core data fundamentals. So I think those two things are what I would suggest anybody getting into it. So build something and really understand data.

Dhaval:
Thank you. Thank you so much Zach. It's pleasure to have you on the show. Looking forward to have you back on the show when we have more things to share and when you publish your next book.

Zach:
We look forward to it. Dhaval
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built a platform to create lifelike characters for immersive experiences | EXCLUSIVE!
14 avril 2023· AI Product Creators
Kylan Gibbs is the CPO and Co-Founder at Inworld AI. He is the former Product Manager at DeepMind, Consultant at Bain and also Co-founder at FlowX. In today’s episode, Kylan Gibbs shares his experience working in AI startups and consulting, as well as his time at DeepMind working on conversational AI and generative models. He emphasizes the importance of iterative processes, adapting to market pressures and user feedback, and the need for creativity in defining good content in the AI space. He advises aspiring product creators to focus on building something that validates their value rather than teaching others before learning. Tune in to hear Kylan's insights and experiences in building Inworld AI and how you can apply these lessons to your own product.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Kylan Gibbs:

• LinkedIn: https://www.linkedin.com/in/kylangibbs

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval: This founder and his team built a platform to create life-like characters for immersive experiences like games or books. In this show, Kylan Gibbs shares his thoughts on generative AI and how to manage products where you create immersive experiences. He also discusses his career journey. Kylen is a former product manager at DeepMind. He was also a consultant at And a co-founder at Flow X. Now he's a C P O and co-founder at In World AI.

Hey Kylan, welcome to the show. Tell us about where you are in your product journey, a little bit about your product as well.

Kylan: Awesome. Yeah. Thank you so much for having me. Super excited to be here. So At InWorld, we're building a creative suite of tools that allow people to build these AI characters and then integrate them into immersive experiences, games, entertainment, enterprise experiences as well. And we started just over a year and a half ago. And basically where we're at now is we have this studio where people can come in and actually craft their characters. We got integrations with things like Unity, unreal, where they can bring them into games, as well as Node for actually bringing these into web experiences. We've also then got sort of this arcade and, basically, which is a way to actually share these characters directly to the web. And then a suite of experiences that we're gonna be releasing this year that are self-produced. So we've got, for example, in world origins to sort of show off the power of the product. And this is all to sort of say that this is kind of crafting that end-to-end user journey of being able to build these characters and then bring them into worlds. And basically where this has all been going is kind of setting the foundations of actually being able to not just create the characters, but, build them into experiences and deploy them scalably. And I think, compared to a lot of the things you're seeing in generative AI right now, like it really is production ready. And we've been focused on that. And so we've already seen a lot of developers starting to churn out games and experiences now that are integrating in world characters. So it's super exciting to kind of see that. So, of course still early on, I'm excited to see where it goes, but already seeing hundreds and thousands of users using it, plus, actually seeing live experiences is pretty magical.

Dhaval: Wow. There's a lot there. So let me just quickly ask you a follow-up question on how you support game creators. Is that right? Or do you support all kinds of creators, like writers, novelists, or just focus on gaming experiences at this?

Kylan: We're supporting really any type of creator. So ultimately, we have users who are, for example, very, well-known science fiction authors who are using their characters to iterate on experiences and potentially write new books with, we have people actually building like AAA games and these types of experiences where you're building multiple characters and integrating them into worlds. We have entertainment companies who, you imagine integrating these into parks or like live experiences where you're actually interacting with. And then we also have enterprises who are using these for things like, brand representation, corporate training. So really we're open to any types of creators. Of course, we're really focused on narrative oriented, immersive experiences, and our product is best built for that. So, you know games, narrative, entertainment, you can think about the adaptation of movies and IP to these experiences. And that's really what we're targeting. But we're really open to any types of creators, and we're finding new ones every day.

Dhaval: Yeah, I picked up a keyword there. Narrative oriented, immersive experiences that could be representing multiple customer segments. What is your product journey like? How do you go about defining your product capabilities with such a broad range, range that you know you could be?

Kylan: I think that, so abstracting out of our specific use case for a second, like I think when you're building a developer tool, you always have two customers. You have to think, keep in mind, one is your creators, your developers, and the other is your end users. And so for us, our creator journey is really people coming in. They have ideas either their building an existing experience or they're ideating on a new one and they're using the studio and our characters to basically iterate on that and then integrate it through things like Unity and Unreal to actually bring those to users. And so when we think about success in that, it's basically: is this person able to create the character that they love and like, and kinda ultimately represents the vision that they have and fulfills the purpose? And then are they able to integrate that and deploy that successfully. Then you have the actual end users, which are the people actually interacting with the characters that the developers built. And that is really like, is the interaction enjoyable? Is the person you know staying around to talk with this character? Are they finding out what they need to progress through this experience? You can imagine throughout a game you have someone that's a guide or shopkeeper and they have to fulfill a particular role. Are they doing that successfully? And so you really have to balance the two of those. I think that's true for most developer tools, but it's kind of unique here because, ultimately there's a key part here, which is like, it's the generated content. So it's almost like we are allowing developers to create characters that are fulfilling the wants that they have for the users in the end? And so it's always a little bit of an art and a science.

Dhaval: Yeah, there is a lot of art there, especially when it comes. The narrative, the experience, right? The character creation and then narrative and the experience. What are some of the ways you create these characters that actually immerse themselves and are conducive to the experience? Is there a difference in product creation? How do you actually adapt to the narrative or the experience that the creator is trying to have?

Kylan: Yeah, I guess there's two points, which is like design time and run time here. So at design time, you could take for example, Arah LA a large language model and generate sort of responses that are aligned with a particular character. But getting them to do that reliably and stick to a story and actually fulfill goals and actions is very difficult. So we actually allow users to come in and they can specify, for example, scenes for their story, and then the characters will actually stick to basically, the kind of motivations and goals that they have for that specific scene. Then we allow them to specify, what is this character supposed to accomplish in this? How are they supposed to speak? What types of things in the world might they be reacting to? And all of that sort of is going towards controlling and biasing the character behavior in a specific moment within a specific story. So they're fulfilling what they're supposed to. And that's all kind of the design side. When you're actually interacting with the character. Then we introduce, for example, emotions. So the characters actually react with emotionality. We have voices that also integrate that emotionality, so the characters can actually, you know, you could hear when you upset a character, for example, and you can react to that. We then control gestures and animations. The characters can actually react to you, or you can ask them to perform a particular action and they can actually act on that. And so, that ability to actually have, and we'll be releasing this soon, and we have a new system where you can actually, for example, give a character a goal and a series of actions that they can use to pursue that goal. And they'll autonomously pursue the goal until they've accomplished it, which is like a pretty magical thing if you think about the ability to actually create this living thing that's kind of, pursuing goals and motivations. But of course all of that comes back to the ability to actually drive this story or narrative forward, whether it's something like Assassins Creed or Far Cry or one of these games. Or you can think about even in an enterprise experience where you're trying to usher a user forward through sort of a brand experience. All of it is really the key point is that the characters are filling a specific purpose in some broader experience. Yeah. And so we've got a huge amount of controls that are available to enable.

Dhaval: That's very interesting. You mentioned that you are able to add emotionality to a character. Wow. That blows my mind. Is there anything more you can share about how you go about doing that? I'm personally interested in that because a few years ago I created a startup to inject emotionality in writing. And that time there were no elements or any of that available, and it was pretty challenging, but I would love to hear how it's done in 20 23, 20 24.

Kylan: Yeah, so right now our system is really composed , so when you as a user are interacting with a character, you're taking all of that input. So whether it's audio, visual sort of event side, so you imagine a you unreal, you have these events that happen and all of that get fed through basically a series of, you can think of 'em heuristics models or models that are representing different parts of function. So for example, you have a goal-oriented module, you have emotions, all these kinds of things. And so emotions are literally kind of, you can think about as an engine. Is taking in the character's current state, taking in all those inputs from the user, updating the current emotional state, and then that feeds out emotion as basically one input to the overall behavior of the character. So then they'll respond to you incorporating that emotion into their conversation or their visual representation, all these kinds of things. And so literally the emotion engine is effectively this module that takes in inputs from different. You know, updates basically this emotion, which is represented as a continuous state, and then outpays, a discreet state, which is like, happy, sad, frustrated and distressed, these kinds of things.

Dhaval: Okay. We are gonna switch up the gears a little bit. I think you have, this is not your first AI startup. I think you have created many AI startups in the past. Tell us about your journey leading up to this involved experience. And what were the learning lessons that you had to share with aspiring product creators? What are some of the, so first, set the context and then tell us your learning lessons.

Kylan: Yeah, so my background was pretty, atypical. I didn't set my trajectory to become an AI product leader. It all really started with like, I was actually starting to go into politics, and then realized that nor to aggregate people's interests and insights and actually understand what they really needed. You couldn't do that manually. You need to do something like data science or machine learning. And so I actually started using it, for that purpose as basically working with the city to understand, you know, population interests. And then, my first steps into the startup world where after I finished my masters, I worked with a few friends to basically create a company where we were applying machine vision to CCTV, which is like security cameras. And basically extracting out information, and the idea there was to be able to feed that then into traffic systems. You could actually optimize traffic systems in real time in cities in the uk. We were working with Singapore and the UK at the time. And that was a super interesting first learning experience because of course it was like, going through the, straight out of university doing the startup life was very different than it was this time around. But it taught me a lot about, I think like that, the importance of. I think one, accepting the technical complexity of what you're working with, which I think can be very Dramatically underestimated, and really being realistic with yourself about what you can do and how much you can accomplish, without a team. And then also thinking a bit more scalably, for example, working with the public sector is a big challenge, so maybe it's not the best place to start out a machine learning product and these kinds of, And then I left that and went to consulting. So I worked at Banning Company and helped set up machine learning practice there and worked as a data scientist cross consultant, which was very interesting and got exposure to a lot of industries. I think the key thing was at that point, which is several years ago now. Getting in uptake from a lot of these large organizations, especially around innovative things like machine learning at that time especially, was very challenging. And also the way that consulting firms or large corporate work sometimes is amenable to building really hardcore tech. Whereas, you're basically focused on churning out client projects versus actually building something foundational. And so you have to put the time into that. Then. I was lucky enough to meet someone who was moving over to DeepMind while I was at Bain and transitioned over to DeepMind. While at DeepMind, I first started working, helping set up the research strategy and doing, you know, basically we were doing a meta-analysis of research across, you know, the industry to inform like research directions. And then around that time, sort of G P T three came out and a lot of these sort of large language model efforts came out. And I was working then as the product lead for conversational AI and generative models, as they pertain to sort of your language tasks and these kinds of things at DeepMind. And so I was working across Google, to basically integrate DeepMind tech into Google products. So it's super interesting, you know, basically learning about, even within Google, where are the challenges of applying generative models and like, you know, this more advanced AI, which is a lot more challenges than people would probably anticipate from the outset. And then that also informed, how I thought about, okay, the benefits of being in a large company versus being in a startup or working in product in a small startup where you have a lot more capacity for experimentation when you launch something, it doesn't have to work for a billion users from day one. You have that capacity to work with a smaller group of users to refine a product and then actually be able to scale that up. So while I was there, I met Mike, who is our current co-founder and c T o, who is leading AI and ML at Cloud conversational ai. And then Ilia, who previously founded Dialogue Flow or API at ai, which became dialogue flow when acquired by Google. And so that's how Inworld was born. Yeah. And then I think there's of course, this has been sort of a rocket ship, so there's been a lot of journey learning along last year, so I can pause there, but I'm keen to share as well. But a bit about what we've learned in the last year or so.

Dhaval: Wow. There is. Five careers in that one career. Thank you for sharing that. What do you recommend, what are some of the things that you recommend avoiding new brand new pm just thinking of becoming a product person or a product creator. What are the things they should avoid doing? Absolutely. If they want to be thriving in the aI space.

Kylan: Yeah, I would say, to boil it down, I would say like, don't philosophize. What I see a lot of people doing, I think because they're mimicking what, so when people have been successful towards the end of their career, they start to, philosophies and they go on Twitter and they go on, post blog posts about all their philosophies and learnings. And so what I see people doing at the beginning of their career, Is basically starting out with that versus actually building something that validates what they've actually done and proves that they're valuable. So don't start by trying to teach people lessons before you've actually learned anything. But that also translates, I think, into actually how you build products. And I think it's increasingly where. The differences between even software development as it existed 15, 20 years ago, and what we're doing now with a lot more AI and ML is that things need to be a lot more iterative because outcomes aren't as deterministic. So you don't know, for example, if a feature is going to be successful. One, because the actual development of a new machine learning model doesn't directly translate into something specific. It's not like you're adding a button to a page that you know has some outcome for a user. It may or may not improve, increase, performance and quality in some other ways. And so what I frankly learned the hard way was, you shouldn't set out, for example, like long-term visions or roadmaps. This is like a very classic thing that I see being told to a lot of early PMs is like, build out your roadmap. Build out your five year roadmap, break it down to one year, and then six. I think that the process is a lot less sexy, but is ultimately far more important. And I see so many PMs, like across their career base, like, oh no, I don't wanna be involved in any organizational stuff. Like I don't want to be involved in the Jira tickets or these kinds of things. But I think a really good PM. Is basically able to take the vision that they have in their head and then turn that into basically a machine or a process that naturally pushes people towards the direction that you're working in, but is able to adapt to things like the market pressures, the reactions, the user feedback. Because if you basically plant a point on the map that you're working towards, you're gonna be so biased by every time that you perturb ba from that point that I think you're not gonna end up in the optimal state. Whereas what you really want to do, and this is coming from the AI ML world, in AI and ml, you have this idea of gradient descent, which is basically you're doing an optimization problem where you're trying to find like the lowest point in a space, I think of good product building as basically inverse gradient descent. So you're trying to always find the highest point. And if you say, I'm gonna, getting to the top of that hill is my goal. Hands down, that's all I'm gonna do. You might be missing the fact that there's like a mountain on the other side of that. And so you should build yourself, you should build your product, you should build the organization and the process around your product all to be able to constantly take in that feedback and shift dramatically. So for example, I've now set up our organization to be more focused on smaller iterative cycles. So, breaking down, you have like larger term, longer term efforts, which you know, are important and thematic. And then you break those down into two week increments. So it's like standard, more standard sprint planning. But the challenges that With A N M L products, it's a bit more experimentation. So it's not like you're necessarily saying we're gonna have this launch and have this impact in two weeks. I'm gonna run this experiment within the next two weeks because I think it might have this outcome, and if it's successful, we'll build on that. And so it's really like taking that long-term vision that you have, setting the metrics, setting the sort of general directions that you have, setting your feedback channels from. And then being able to constantly turn that into experiments that you wanna run, both with developing hardcore tech. If you're doing a deep tech startup or, turning that into, user research or user experiments. You've been doing something that's more front done based.

Dhaval: Very interesting analogy there with gradient descent. I often use the same analogy, but I use a different term for it. Global maxima, find your global maxima. Yeah, I love that. Let me ask you this question. Where do you. This is going, this whole, generator AI especially. And then if you wanna narrow down to your specific space, I would love to hear that.

Kylan: Yeah, so I, I generally think where we're moving towards is, In the same way that you had basic automation of things like databases and things over the last 20, 30, 40 years. I think we're moving to a place where a lot of like hardcore knowledge tasks are going to be automated. And ultimately, this means things like you imagine as a PM creating a P R D, you can probably generate a pretty good PRD of chat GPT, barely better than most starting PMs to be honest. And, you could probably, you can create, there's like slide generators now for people who are in consulting. There's a lot of these things. So knowledge tasks are becoming more automated and I think it puts a lot more pressure on ideas and creativity as like the actual gold of what actually humans contribute to this whole process. And so what I see happening over the next 10 years is, A lot of the things that people held as basically their like unique value are going to be eroded as what happens like with any technology innovation. But what that means is, so let's say that was only 1% of the population, now those skills are accessible to 80% of the population. So the overall value created by everyone goes up dramatically. But it takes away the feeling, the niche, sense of value that small group once had. And so what I think needs to happen is for people who are very, you, Highly technical knowledge oriented that feel like this is impacting their work. The question is not "oh no, how do I react to these things being taken over?" It's like, how do I also now become an expert in taking advantage of these tools? So, for example, in our case, so you have. Amazing game developers, narrative designers, writers who are creating video games. And it's not the fact that creating these AI characters who are able to basically generate dialogue is gonna take away from that. It now means that a single writer could create a hundred characters and create all the dialogue for them in the time that it would've taken them before to create one character. And what this really means, I think, is you're just gonna have an explosion of content overall. So both in terms of games and things that are directly impacted by us, but of course there's also the image generation and text generation, and I see music generation happening now, and so. I think we're about to enter an era for probably the next five years of just like a content explosion where there's just gonna be so much of everything and a lot of it's gonna be complete noise. And so I think the key thing for like those people who want to continue to be leaders and product leaders in this space is not what can you create, but. How can you filter out the noise? So how can you ultimately define what good is? Because it's gonna be a lot easier to create something. And so I think that like over the last, while we've seen the benefits of people who are just able to build some hacky prototype. Building a hacky prototype is gonna get really easy. So now you've gotta figure out like, what are you actually creating that is completely unique and like applying some creative sense, not just necessarily some technical or general sense, or applying some framework because guess what? All these generative models have now taken in that framework and know it, so you don't really have anything special. So you have to come up with something actually new now. And I think that's where it's all going is like a content explosion that biases us to really thinking about what the content is that we want to create, not just the fact that we can create content.

Dhaval: Yeah. I love that. Thank you so much Kylen. Thank you for making time for. Really appreciate it. Looking forward to having you back on the show once you have a few more lessons to share with us.

Kylan: For sure. Thank you so much. Yeah.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
Episodes manquant?

Cliquez ici pour raffraichir la page manuellement.
How this founder built a personal language model to mimic your behavior, knowledge, and style.
5 avril 2023· AI Product Creators
Suman Kanuganti is the Co-Founder and CEO at Personal.ai. Previous to Personal.ai, Suman also founded Aira. Suman holds his BE degree in Engineering, MS in Robotics, MBA in Entrepreneurship, and ten patents in emerging technologies as well. Suman also founded Aira. Personal AI is a GPT implementation designed to mimic an individual’s behavior and to speak like them. In today’s episode, Suman shares his goal is to create AI systems that understand and replicate users' communication patterns and cognitive abilities. Suman emphasizes that their language model has time awareness, allowing it to adapt its knowledge base depending on the user's age or point in their timeline. Suman also shares his some learning lessons for AI product creators.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Suman Kanuganti:

• LinkedIn: https://www.linkedin.com/in/kanugantisuman/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
So today we have Suman Kanuganti on the show. Sunman is the co-founder of personal AI, a product that he will talk about and, we'll learn a lot more about where his learning lessons bear, what his learning lessons bear and all that other stuff. So yeah, let's get started. Suman, would you mind introducing yourself and tell us a little bit about your product, AI product?

Suman Kanuganti:
Sure Dhaval. Thanks for having me. I'm Suman Kanuganti. My background is in engineering. Over 10 years ago I started creating companies. This is my second company. Previously I built a company called Aira, A-I-R-A. My philosophy always has been how do use technology kind, solve, hard human problems. Aira was about using technology to fill the gap of missing visual information for people who are visually challenged, such as blind and low vision. And personal AI is about augmenting people's mind where memory, cognition, and our time is limited. And we would want to augment that using technology by creating a personal language model of every individual that essentially learns to behave and act and learn to be you. So that's a little bit of my background and who I am.

Dhaval:
Wonderful. Thank you, Suman. Thank you for that introduction. Tell us a little bit about what personal AI's narrative is like. What is the, who is your target customer? What kind of a problem, specific problem you are solving with that product? And what is the overall narrative? How is it different from your competitors?

Suman Kanuganti:
Yeah, totally. As individual people, like on a day-to-day basis, we create and consume a lot of information. Like, you know, a lot of experiences have a lot of conversations. But obviously, 80% of that is lost or forgotten. Our goal with personal AI is to be able to create a model. That actually learns your knowledge, your style, and your voice, more or less like to be a digital version of you. Imagine being able to surface relevant pieces of your own knowledge, on demand whenever you need it. Or imagine, you having conversations in a, chat or text message. Are people talking to you? Where relevant pieces of information when we start facing as you speak. So our intention is to be able to augment, humans with an extension of their own mind. Because one cognition is limited and then two time is also limited. You mentioned about the target market. Our goal is to actually go after everyday consumers. Our intention is to have everybody have their own personal AI. That is trusted by them. The data belongs to them. The model gets trained over a period of time and innovate kind of grows alongside with you. Unlike, public or general models that exists such as open AI, such as, Google or Alexa, which is mostly like trained on public internet of data. Personal AI is a unique model that also uses similar architecture such as GPT, but actually trains on individual person's data. And it does so stylistically, relevantly authentically to replicate as you would. So we are trying to essentially like replicate, your thought process and your mind and give you an extension of your.

Dhaval:
Wow. So it is adding the stylistic and tonality and the personal attributes to your, to the replica that you are building for someone which is not there in current GPT or any of those products, right? So there's a bias that. GPT has very confident answers, but doesn't necessarily align with your style or may not align with the way you communicate. And what you are saying is that not only does personal AI helps you do that, but also creates the model in the first place using your personal attributes , did I get that right Simon?

Suman Kanuganti:
Yeah, totally. So you will create what we refer to as a memory stack, which is essentially taking all your unstructured data that you ever have in your digital world. Let's say you're having conversations online, you are texting with people. You probably have written a bunch of different knowledge pieces out there. And then we create this memory stack, which is essentially like a digital representation of your memory vault. In other words, we basically break down this idea of structuring your data into these blocks that is associated with time. And imagine over a period of time, as you create an as you learn. Your AI technically would also be training alongside with you. So it's kind of how. Conceptually we have architecture system.

Dhaval:
Got it. Now, there are, if you were to dive a little bit into your products architecture or the core engine, the AI, since the audience of this show are the people who aspire to either create an AI Product. Or wanna add AI to their existing product, for their knowledge. If you were to share a little bit without getting into the confidential details about what does a product stack, what does it take to build memory stack? What does it take to build a knowledge replica, a knowledge brain summons, human brain, and human personality into something that you were doing, like what do you call that thing? The entity that.

Suman Kanuganti:
Yeah, I'll try to provide answers and then I'll try to provide some contrast technologies that exist out there so that we can wrap our heads around. The first thing is at the core, we are essentially an AI first company. We built an algorithm called. Personal language model. So we call it personal language model. This is in contrast or like kind of opposite in concept to a large language model. If you think of a large language model such as GPT 3 or any other open Language models that exist out there close to around like one 70 billion parameters in our case. Our language models around one 40 million parameters. It revolves around at the core individual's data and not public's data. And you can keep on adding the data to your model so that way it gets more sophisticated in regards to the purposes of your mind. You can go abroad and you can also like, go deep into specific topics itself. So yeah, at the core we build this personal language model for every individual to essentially like mimic the behavior, knowledge, and style of an individual person. And the transformer that we have developed, we call it generative Grounded Transformer. And if you think about like GPT as a generative pre-train transformer, the subtle difference of our transformer is that it is grounded in the personal data of you. And whenever I refer to personal data, is nothing bad. The memory is tag that I was explaining earlier, So every AI response that your personal language model actually generates, it has an attribution, and the attribution is nothing but attribution Back to the data or what"s? Data elements of what memories in within your stack is responsible for creating a particular response? One of the challenges for large language models is that there is no attribution, primarily because it is driven by aggregation of the data. And there is quite extensive anonymization that is involved. So technically, you cannot create that attribution and it's extremely hard to create that attribution. And our goal is exactly the opposite. We would want to have. That attribution, we want to have the ownership and we want to create that value to every individual consumer by creating their own individual model, at the foundation, it's personal language model.

Dhaval:
So this generated you, grounded trained model that you are referring to is that, Adaptive with human changes, human behavior changes over the lifespan of the user.

Suman Kanuganti:
Exactly. The transformer also has a sense of time. For example, let's say, if you're talking about AI maybe three years ago, How you refer to your transformer, how you refer to your technology, maybe different from your latest and greatest creations or thought process around your ai. So it has a time to decay component. So when you are indeed chatting with your own AI, it normally anchors around the latest and the greatest thought process of. How you would respond. However, let's say if you indeed are contextually trying to fit something from the past that happened like 10 years ago Then, we are probably talking about the first autonomous car, right? And my experience with the first autonomous car, then it will. It'll go back in time and be able to fetch that response for you. So kind of like designed to work very similar or akin to how a human mind would function. You can think about potentially being able to drop your AI at a certain period in time. Given these are small models and given as the data is going in on a day-to-day basis, there is a new version of the model. We are technically able to time travel. Your model, like let's say 2, 2020, and then when you are having the conversation, it would like replicate the information density as if you were to be functioning at that time. If that makes sense.

Dhaval:
That makes sense. Yeah. So with this language model the way you communicate is it's having a time sensitivity to it. It has a time awareness to it. So it, if you go back to your younger age, you would have smaller, it would have smaller set of parameters. You look at. Depending on, the size or depending on the scope, it may change depending on at what point in the timeline you put it. So this is, what are the use cases? What are the use cases of something like this or an average consumer? What would they benefit?

Suman Kanuganti:
yeah. I'll give you a few different use cases, but I'll also tell you the focus of the use case that we are going after. The few different use cases could be. Anywhere from simply being able to remember everything that happens in your life and be able to recall. It's almost like a data store where you're not searching for the data, but you're actually interacting with yourself. And being able to, recall pieces of information and facts as well. So that's fine. Our personal language models also has different capabilities, meaning as increase your stack size and get more data in it moves from being able to simply answer questions to being able to generate content for you and being able to have a conversation like you as well. So that kind of unlocks into. Being able to create or generate responses or draft emails for you, write tweets for you and even large form content as well. And over time it could technically be a conversational mind that exists on the internet for anybody to interact 24/7. And an asset that could label potentially on the internet forever for your future generations to even have a conversation as well. So it's designed to be an digital asset that essentially grows over a bit of time and they own it. The use case that we are going after is to be able to draft responses in a human communicates with other humans currently in multiple different platforms. And one of the downsides is, everything is kind of lost and you are pushing the data to large, big tech companies. And our intention is to create a, personal AI chat system where people would communicate. But every conversation that people communicate in within that system is trained upon. So if they, give a piece of knowledge or information once. It'll be useful or re useful at later bit of time, given the particular context again and again. So think about as like a automatically drafting solutions almost all the time. And you can potentially put your AI in a co-pilot mode or a auto-pilot mode that is communicating on your behalf with other people if you would choose to all the time. So it saves time, but it also saves, the idea of augmenting your cognitive capacity. Because it's very hard to process all the information all the time that is needed.

Dhaval:
Wonderful. Thank you Suman. One, couple of quick questions follow up and then we'll wrap is what are some of the learning lessons you had from your product journey? That's a second. The more important question is where are you in your product journey. Have you launched your MVP. How many users you have? Where are you in the stage of the company? If you could share that and then if you could share any learning lessons in this journey, any big learning lessons you had for other AI product creators?

Suman Kanuganti:
Yeah. We had a little less than three year old company. We've been focusing mostly on creating our language models and testing with a few groups of people and experimenting to figure out like what our market is. So we would be launching our. The first version of the product in March there is a version of the product that people would come in and train their AI to, essentially remember and recall information and train model over a period of time. But technically the first launch, the personal AI chat application. We call it personal AI to that, or we'll be launching in March. The learnings are interesting, so the learning. First of all, like, almost two and a half or two years ago, even when we talk about personal AI and creating a digital version of your mind it was almost like too hard to delay. let's think about it. I am not sure if it is there. And then the GPT three came along, and when GPT three came along in 2021 it was good for AI awareness. And then it was taught as a almost like a marketing tool because it was really good at content generat. There are good number of, startups that are evolved who are building on top of GPT three which is content generation tools. We were heads down essentially finishing our development on the tech as well as on the product to create this personal language model. It has the generated capabilities but it also has conversational capabilities. One of the insights was we went after like content in Generation. But the most interesting thing is large language models are so great content generation. The content creation normally happens with new content now, I found people who would want to create content from the existing knowledge. The appetite is very low or maybe it's like a different use case. So our, purpose of the personal AI is more valuable where. The existing knowledge is more valuable, the existing interactions and experiences are more valuable. And that's where we landed on identifying our PMF, like where it's everyday consumer application for being able to augment your conversations and communications and make it more human to human and stuff like human to ai because human to Ai. Is a very, like a bot conversation. And our goal is to create this ai be like you and not necessarily talk to your bot. So there are like several nuances around how we create these experiences, which is which was exciting in a way. So yeah, so essentially where we are driving towards right now is Chat GPT came along. The awareness around the AI essentially being able to chat. Is more acceptable because people saw the taste of what is possible and it is great because it's all trained on public data. So when we talk about personal ai, which is essentially chat GPT, if you will, but for personal data the promise is very exciting and there is awareness and there is accept. So essentially here we are basically going to the market, creating a, not even creating, basically launching our personal AI chat application, which is in of development for past one year. And excited to give it to everyday consumer and people, and everybody will have their own personal AI that will essentially grow with them and it'll be theirs, right? It can make money for you. It can live in the cloud. While you are sleeping, it can do work. So yeah, that's where things are headed to.

Dhaval:
What are you most excited about for personal AI in 2023?

Suman Kanuganti:
Most excitement is this. We've been at it for a while now, and finally things are coming together. What that means is we are finally able to kind a fulfill this idea of democratizing AI. Where everybody will have access to their own AI. And we are also excited about this core idea of access between people. If you think about what does it mean to be able to have access to the loud ones that you know potentially passed away are people who has like knowledgeable but we do not have access to. So there is a lot more, interaction gap that exists. Information gap may not exist because internet exists. So I think personal AI is the next level of, almost like Internet 2.O that will kind of unlock this exchange of information in a much more meaningful yeah.

Dhaval: And you say that AI, generative AI is over a trillion dollar. Where does personal AI fit in this market space, and how much of it do you think belongs to personal ai?

Suman Kanuganti:
I think it fits in the everyday consumer market space. it would likely be sitting next to, any of the Google Assistance services. We are more focused around day consumers to let say, you know, your personally would be living in your mobile phones. So communication like human to human communication and establishing those connections with and increasing access between people and reducing basically burden of being able to remember, recall and generate response. So I think it's somewhere between a communication augmentation tool within the consumer space and not necessarily not necessarily in the customer support are business engagement where general AI are pretty good at. So we really want to tap into this idea of, personal nature of personal AI.

Dhaval:
Now if I were to use personal AI in my work communication, is that part of your vision or is that considered professional work and business AI.

Suman Kanuganti:
You could nothing's talking about. And then at the end of the day, I think every company that starts with consumer application or consumer focus eventually penetrates into the business markets. So we want to build this like ground up rather than going after like existing like business application data and building the models over there because there are technologies who does that really well, right? We don't have to do it. There are not a lot of technologies who are actually like, really good at working on like smaller amounts of data and people's data. So eventually if you would want to use it for work, there is nothing stopping. In fact, there are people who use it for like specific projects, the conversation and augment their conversations, within the professional setting. But I think when I talk about like everyday consumer, we are essentially not going to start off with aggregating of the data in a business setting. We are going to start off with a personal data and individual data starting off with an individual consumer.

Dhaval:
Got it. Well, fascinating conversation soon. Thank you so much for making time for. Chatting with me and helping other AI product creators cut their learning curve. I really appreciate that. Looking forward to chatting with you again once your product is launched and you have a few more things to share with us. In the meantime, I wish you all the best with your launch coming up here and we'll be keeping up with your news.

Suman Kanuganti: Thank you Dhaval.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
What happens when a PhD Professor in Analytics launches an AI Writing Product?
4 avril 2023· AI Product Creators
Martin Pichlmair is the CEO of Write with LAIKA, Associate Professor at ITU Copenhagen and Co-founder of Broken Rules. He Holds an PhD degree (Department of Informatics) in Vienna University of Technology. In today’s episode, Martin explains that LAIKA is designed to make AI-generated writing more accessible and user-friendly, with the AI and the user working in a tight interactive loop. Martin highlights that their product uses a "no-prompt" system, which means users don't need to be skilled in prompt engineering to get meaningful results from the AI. Instead, the software handles most of the prompt engineering behind the scenes, making it easier for users to interact with the AI. Tune in to hear Martin's insights and experiences in building LAIKA and how you can apply these lessons to your own product.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Martin Pichlmair:

• LinkedIn: https://www.linkedin.com/in/martinpi/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
welcome to the call, Martin. Thank you for joining. Tell us a little bit about your product.

Martin Pichlmair:
Okay, so I'm Martin. I'm the CEO of Write with LAIKA. And our product is a kind of creative writing tool that is using large language models, in our case, quite small, large language models to, support writers when they get stuck or when they need more text, that is influenced by their previous writing.

Dhaval:
Wow. Okay. So when the writers get stuck or when they. Interested in continuing with the style and the tone of their previous work. They can use your product including the contents, the storyline, or anything along those lines.

Martin Pichlmair:
Yes. How LAIKA works is that you you upload existing writing. You have when you get, for example, stuck in a murder mystery because you don't know who the murder is. Funnily, we had that case twice already with users. And then you upload what you have written before and our, system fine tunes a language model With your text and then you can prompt the model to continue writing in your voice, in your using your characters. You mentioned using scenes you have been writing about in the past and very much sounding like you. Now you can do that with your own text. Or with the text of famous writers, we have, for example Dostoevsky in there and Jane Austen in there. And a lot of, all of them, of course, dead and out of copyright writers that you can also collaborate with in a similar way by asking them how they would continue a sentence, for example.

Dhaval:
Wow. So it has memory and context as well as style and the personalization built into it. So is that. Large language model that's very different from Chat GPT 3, which would spit out very confident phrases very long phrases. But they're also having the same style. Is that, how is that different from the large language models? You said that you have used large language models or you have you built on top of them or like, help us a little bit on how have you built this.

Martin Pichlmair:
So we've built this on very small, large language models. They're still in the same architecture and come from the same family, but they're very small because that gives us the ability to fine tune them very quickly. It takes like five minutes. If you upload , a half done book, for example, takes five minutes and you get your own, we call them brains because that's a nice metaphor. Your own brain based on your writing to interact with. Now, of course it has an understanding of the context, but it's not always super, like it doesn't have an actual understanding. It can just play with probabilities of words, just like all of those language models do.

Dhaval:
Wow. Very cool. Let's dive a little bit into your product journey, is this your first startup? Is this your first AI product? Tell us a little bit about your background, Martin.

Martin Pichlmair:
So I have a weird background. I did a PhD in computer science originally at the University of Vienna, at the tech university, and then worked in academia for a couple of years. I got a little bit, I don't know I wouldn't say bored, but I wanted to do something differently. So I started a video game company and then after a year started another video game company because the first one didn't work out. It didn't work out, but it also didn't not work out. It was fine. It was just not meant to be a longer existing thing. The second one actually is still around. It's called Broken Rules and makes awesome in the games. But I'm not involved anymore because I decided at some point to go back into academia. So that's where I spent the last seven years until last year where I just realized with my partner, That we have a huge connection between what I was doing in research, which was using generative AI to create systems for video games and her background, which is writing for video games. So we sat down and, uh, started workshops during the Covid Pandemic when everyone was sitting at home. We started online workshops where we introduced writers. To, the newest possibilities in language models, using very, very clunky tools at that point. And after three or so of those workshops, we realized the workshops are always poked out, but it's really hard to work with the tools that are there. So we decided we have to make our own tool, and that became a research project that was funded by the Danish state, uh, in the beginning. With the intention of turning it into a product. And since last November, we founded a company and turned it into an actual university spin off that is based on yeah, research that is now working on a product that we will commercialize within the next month

Dhaval:
Very interesting background. You do have like a very traditional computer science background, making you very competent in this area. Right. So, quick question. You mentioned that you launched this in November, but you haven't commercialized. It doesn't mean that the product has not launched yet.

Martin Pichlmair:
Yeah. We have a wait list and we have, a data with, nearly 2000 users. So there are a lot of people using it every day, but, it's not a commercially launched yet. We're still only free for select users.

Dhaval:
Very cool. Is this. Is this a self-funded or have you bootstrapped this whole thing? Are you intending to, or have you raised capital? And are you intending to raise capital as you move forward?

Martin Pichlmair:
Well, we got some funding from Danish State again. The program that we're in that funded turning research into a product last year that was still in the context of my university has a follow up program that funds your salary basically. So we are kind of weirdly half bootstrapped. We have no investor, but we have, our salaries covered by the Danish State but a very low salary. But still, it's good enough to, know that we'll be, we'll be around for another year at least while this funding runs. we're looking for investment in the moment. It's, we are talking to a lot of VCs. It is. It just takes a while it seems.

Dhaval:
Yeah. how is that playing out in this current market? Like how is that, can you give us have you done this before? And if you have, like how is it compared to the current market, if you can speak to that.

Martin Pichlmair:
So I haven't done it before, but a good friend of mine has a very similar company, actually a very different company, but also an AI company also in Denmark. And he also has an academic background. It's otherwise very, very different because it's B2B and started out much bigger than we are. But it looks like they, the climate they saw two years ago is very different to what we have now. So I'm getting all my tips from him and half of them don't work anymore. The climate is not good in the moment. Even in the hype space of creative ai, there is a lot of chicken egg problem, happening in the sense that investors wanna see. They actually want pay traction very often .They want to see some pay traction or immense numbers in weightless users or something like they wanna have proof of actual viability very early on. But it's such a new area that you are actually creating a market. So it's very hard to say where this whole journey is going because the whole, like AI is not super new. But generative AI is really something that is only a thing since like a year or so. It's very hard to say where the journey goes in the moment and, like it could all still just be overhyped. Then I would understand the need for having paid traction, but it could also be that we are just opening, creating a completely new market here, and then a little bit of trust would be nicer than having to prove things too early.

Dhaval:
Yeah. Yeah. Where are you in the product stage in terms of product development? Are you close to? I know you mentioned you were ready to launch in a few months. Are you close to finishing your product development? Is is that almost there? Like, if you can share that. Cause my follow on question's gonna be on, what were your top learning lessons around creating an AI generated product? What was that experience? What was the top learning lesson there?

Martin Pichlmair:
So I think the whole idea of finishing a product is not really how it works with software as a service anyway, but especially in this extremely fast moving area of, AI in general, but especially generative AI where new technologies come out on a sometimes weekly basis. There's a lot of competition, but there is also just a lot of speed of development. I don't think we will ever be at the moment where we say, now we are done with this. Instead, what we are, where we are trying to get is to a point where we can say, this is our 1.0 version and we hope to be there in the month actually. And, from then on we of course continue building. and one of the main challenges to my surprise actually, and maybe that was the main learnings, is since what we're working with is so new, it is really, really very hard and crucial to explain to users what we even doing here, because they are very curious, but they have a very hard time understanding these new ways of interacting with the computer. And, it takes. Nearly as much time to make the product as it takes to actually package whatever you're building in a way so that an everyday user can understand what is even going on here. So that's quite challenging and that is something that we also are constantly in the process of figuring out.

Dhaval:
Yeah. One of the biggest challenges for generative AI products is gonna be around user experience and managing that human AI interface. What has been your biggest learning lesson in that area for your product?

Martin Pichlmair:
So we have a nearly no prompt system. So most, in most AI systems, how you interact with what a generative AI systems. If you interact with them, you do some kind of prompt engineering as they call it, so you get better and better at formulating what you wish to get out of the model. More and more precise over time and you just, it's a learning process. You prompt them with text usually, and you just get better at describing what you want in the way the machine understands. which is an interesting interaction, but it's actually surprisingly technical. And our approach is a little bit, since we are all about making this more accessible, these systems is our approach hides that. We are, of course also prompting in the background very much from the user and tries to make sure that you get a higher percentage of. Good results automatically instead of giving you this huge space of freedom that you have with more technical systems. But those systems exist so if someone is technically inclined, they can just use a different system. It's just not what we are making a software for.

Dhaval:
Yeah. You mentioned something about prompt engineering and how does that play out when users are engaging with the product, do they get to like decide exactly what's gonna be the better prompt to ask, or is that something done behind the scenes?

Martin Pichlmair:
Most of it is done behind the scenes. So the base use case of LAIKA is, you upload your materials to train with your voice, and then you just write half a sentence or a full sentence and a question mark, for example, and then you ask the system how it will continue from there. Now, of course, technically you are just prompting. But in practice it feels like you are writing and now, and then you give a little bit of control to the machine and then you override the result immediately and work in this very tight interactive loop with the system. So it's a bit of a, yeah, more. There's a lot of discussion about generative AI in the moment among artists that, uh, somehow do not appreciate the fact that there are so many machine generated images out there and what we are doing is very different because there is no result. That is trust. What comes out of the generative ai, it is always the author with. Their own thoughts remixed by the machine creating something. So it's very much 80% is human authored and what is even made by the machine vanishes, hopefully. So it's a slightly different way of interacting. It's using the same base technology and the same post model base modalities of how you interact with a machine learning model. But it hides them and it makes the loop even tighter than existing systems.

Dhaval:
That's where the magic is, making that loop tighter. Thank you for sharing all your knowledge. Martin. Anything else that you wanna share about the future of your product? Where do you think it's going? Do you have a big vision around where it's gonna be in about 10 years from now?

Martin Pichlmair:
About 10 years. Oh boy. Well, in the end I think that I actually have a big vision how, where it will be in 10 years or where society will be in 10 years. I wrote a blog post about it recently. That's why I have. The thing is I actually think that a lot of interacting, interaction with generative models is not so much about getting results and much more about the experience. So what we are doing is we are interacting with cultural, the cultural heritage of humanity, in a live and, uh, uh, back and forth ways. So we synthesize something that uses images that are up to thousands of years old, like the photos of those images, but still human expression that is hundreds, thousands, and so on, years old, and we synthesize them into something new. And that activity is actually a very interesting activity of interacting with the, what humanity is with all of our culture. And I think it will become just something very normal for us to do. That we're not only interacting with the way a specific historian has recorded our history, but in a much closer, again a tighter loop, interact with past images, past writing, past words, past figures, maybe historical figures and so on in a, and of course, synthesized way, but still very much informed by who humanity was in the past. And I think that just will lead to a different understanding of who we are. And that will maybe take 10 years, but it's gonna be interesting. That is a really big vision. I know.

Dhaval:
Yeah. That's amazing. I, I look forward to get to there as well Martin Write with LAIKA 2000 beta testers, how many people who have signed up for the wait list.

Martin Pichlmair: Well, not many more.

Dhaval:
Not many more. Okay. And you are ready to launch your product in about a month's timeframe and, you have exciting, future ahead. I am looking forward to following your journey. Martin any last thoughts on anyone who wants to start on this journey? AI, product creator who wants to build an AI product. In generative AI space using text as their, as their tool. What would the, your advice be for them? any tools that you would recommend for them to use to get started?

Martin Pichlmair:
I think the important thing is to find a niche, because I think in the very near future, like next year, most likely, Microsoft will trust, integrate generic functions for creating text into all of their products. And then that's just gonna cover 90% of the use cases. So if you want to get started in text, it's like you have to find a very, very, very specific niche, maybe in the business to business sector, to be able to still be around in a few years when that will just be a standard function of word processors. So I think that is my main tip to just. Yeah, keep it, keep it focused.

Dhaval:
Now in terms of execution to keep it focused, would that be custom models? Would that be built in tune with GPT 3 or GPT 4, whatever is the next version? How would you suggest that?

Martin Pichlmair:
I do like our autonomy. We own nearly all of our stack, all of our core stack is owned by us. And that gives us the ability to define our own policies of how people can interact with our system it gives us safety. It makes us independent of the server costs of other people. Of course we are hosting it in the cloud. It is a little bit cheaper also in the moment. So this autonomy is something that works very well for us. Now, I think it's of course, expensive to build a product, using your own stack, but I think there is will just be more and more need for diversification of what you're doing. And we can have more diversification than any product that builds on the vanilla GPT 4 version or 3 or 3.5 or whatever.

Dhaval:
You cannot really have that niche if it's built on top of a vanilla version. Well, thank you so much, Martin. It's been a pleasure having you on the show. I'm looking forward to following your journey,Thank you so much, Martin. Have a great rest of your day.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This former Dropbox engineer built an AI Product to create stylized eCommerce product photos
3 avril 2023· AI Product Creators
Boyang Niu is the Co-Founder of Stylized. He holds Bachelor and Master Degrees in Computer Science from the University of Pennsylvania. In today’s episode, We discussed his journey and vision for the AI-powered e-commerce platform. Boyang shares the long-term vision for Stylized, which is to become an asset-first e-commerce platform, simplifying the process of setting up an online store by taking care of all abstractions, from website building to SEO. Boyang emphasizes the importance of understanding one's strengths, whether it's distribution, core ML, or UX, and iterating quickly to create a high-efficacy product. Tune in to hear Boyang's insights and experiences in building Stylized and how you can apply these lessons to your own business.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Andrew:

• LinkedIn: https://www.linkedin.com/in/boyang-niu/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
This former Dropbox engineer built an AI product to create stylized professional product photos for people running e-commerce shops. Boyang is a founder of Stylized ai. He Figured out a way to leverage depth extraction, AI and 3D rendering to empower e-commerce sellers to transform their phone photos into professional assets for everyday use. One of the biggest lessons I gleaned from this conversation is how he approaches product development by focusing on areas of growth and applying the user mindset to product creation.

Hey, Boyang , talk to us. Tell us about your journey. How did you come to, identify this, space? And, tell us a little bit about you first and then we can talk about the product.

Boyang:
Yeah, absolutely. Great to be here. Dhaval I joined Hive. So Hive was a social media company back in 2015, building like a Twitter clone. We pivoted hard to enterprise computer vision SaaS. And so that's where I first like learned about ImageNet, um, did all this AI stuff built like a trading pipeline and all of this stuff. And then after that, worked in some productivity tools at Dropbox, worked at e-commerce at Square, and then for the latest venture, I like put all those things together. And really it's about targeting the market that I care about, which is e-commerce sellers, which is a huge market, right? You get a lot of customer iterations, a lot of customers to go after. You don't have to really be scared to approach any particular one. And that's a great feeling when you're just getting off the ground, right? Because you need that iteration cycle. And so we, we. Sort of putted around me and my co-founder, looking for ideas that really resonated. And one of the things was like, oh, people want to take photos, right? People, when they sell things, they need photos of the thing. that's where buyer decisions are made, and they pay like 35, $50 per photo professionally for these images that they're putting on Shopify. And so people are coming up to us, they're being like, oh I waited like six weeks for this photo set. For, 250 photos, it cost me 10 K or 5k, whatever. And we're like oh this is interesting, right? Because, there's this new image stuff going on and maybe we could really leverage that, to make this workflow better.So that's really where we came up with the idea for, what is now Stylized, Stylized.Ai. What we're building is professional product photos, for people running e-commerce shops in under 30 second

Dhaval:
oh wow. So tell us a little bit about where your product is at this stage and have you launched, is it in the pre-launch stage? Is it in the wait list, stage, et cetera, et cetera.

Boyang:
Yep. We have launched, we are soft rolling out a launch with this is a prosumer product, right? And so we're really concentrated on the B2C motion of go to market. We're doing like organic seo. we're running a bit of ads on the side. And so this is all just to build up , a brand name and also to get really fast iteration on what the product surface is. So we have soft launch. We have about 200 customers right now. That's growing probably at a rate of, I would say, like 15 to 20 per day. Which is pretty good, right? It's only been like, I think since we opened the beta. To one and a half weeks. So we are pretty happy with that. And I think the goal for us really is to get that distribution and get in front of people into their workflows such that we get embedded. And because no one really knows, like honestly, no one really knows where the AI models are going. And so by in the next three months, someone could do some really magical stuff, right? And we want to be able to put that magic in front of customers. And to do that, you have to have a big audience. So that's our goal, right now

Dhaval:
That's awesome. Focusing on distribution first, that's novel. Most of the founders and product creators, they get their heads down and they start building the product and it. They spend months and months and months before even thinking of the first interaction with the customer. And as you already know, it never goes as per the plan. Right. So what is the main value proposition? What is the main customer pain point that your product solves for?

Boyang:
Yep. Customer pain point. I have a product. I'm trying to sell it. I have a Shopify store. I need good images. Right now my options are I get like a light box set up, which is, they can be pretty complicated, right? I need to set up a photo studio area in my house. I need to take pretty meticulous pictures. I need to then learn Photoshop and edit those just the way I want them. Or I go to a professional studio and get my photos back to me, in a couple weeks. And so that's my blocker right now. I can't get it on Shopify. So what we do is, you. An iPhone image, and as long as it's like pretty good neutral lighting, like anyone can do this. I've done this many times and I'm bad at taking photos. Right. So I do that in my product. In 30 seconds I get a virtual light box. So this is a staged, 3D rendering actually. And the technology is relatively, I wouldn't call it simple, but it pieces together a bunch of existing models, right? To render your product in 3D. And then you get to adjust whatever you want about that rendering such that you know your product is professionally lighted. You get to change all the backdrops as if you are in a product studio or in a photo studio with like different types of materials or Hey, I want this on marble or slate, or all of that stuff. But you get to do that from the comfort of your computer and the iteration time is, on the order, five seconds versus two weeks. so that's what we're going for.

Dhaval:
Wonderful. Yeah. I've been a product owner for e-commerce companies and that finding good stylized photographs of your product has been the biggest game changer. Like experiencing, showing people experiencing the product has been the biggest game changer. So you're solving a real pain point. You said you are in a prosumer space. If you can unpack that a little bit, why is that presumer and not just e-commerce sellers. How do you differentiate?

Boyang:
Yeah. So for us, the biggest differentiation is whether we are b2b, which would be selling to e-commerce platforms. And we've talked about this as well as whether we could go to Shopify and say, Hey we have an api, or we have a third party tool that. You could purchase for your sellers to make their shops more efficient. Whether that's the route or if we want to go directly to the customers themselves. So we see that as more presumer because it's self-serve one, we're just launching to anyone. You can come in, you enter your email, um, you upload a picture and boom, it's there it's free to use. You just get these premium add-ons. And that's how you are introduced to that product at first. So we're calling it prosumer mostly because all of our customers are independent shop owners, and they really get to make the decision about their own product.

Dhaval:
Very cool. Yeah. So your product roadmap could be either build up your distribution, get a lot of, Prosumer e-commerce, use your product, and then become down the line, become this extension or plugin for all the e-commerce outlets that are out there. Is that something you're thinking of?

Boyang:
Yes. I guess I won't go into too much detail, but we do have a Shopify extension that's coming out soon. As I said, we're focused really on hitting that distribution and just nailing it, getting as many customers as possible. And one of the, one of the benefits here is like we're solving one. Very individual problem, right? So it's, Hey, I need a photo. There's a very clear input, like I take a photo, there's a very clear output, I get a better photo back, right? That takes 30 seconds, 15 seconds, whatever it might be. We're solving that pain point. So it's very easy to get in front of people and say, Hey, look. This is what you're getting from us. and it's easy to onboard. And then from there, I think the strategy here is if we get many customers, we can start building up, catalog extensions. We can say, Hey, put your entire store catalog with us, like we will optimize it. Or you get better photos through all of your store. And then we can really start to leverage all of the newer AI things that we see coming out in the future. So if one day there are, very good AI models for just creating catalog webpages, we could leverage that. We could then let you know our customers create their own webpages directly from our surface. And that is really the expansion route that we're foreseeing. It's like an asset first, website.

Dhaval:
Very cool. So we have talked about your journey. We have talked about your product. We have talked about your potential future roadmap. What I would love to dive into now is unpacking the product stack a little bit for people, for product managers or product owners who want to either make an AI product or wanna add AI to their product. What was your journey like when it comes to building this AI product, what was the stack like? Anything you can share that would really be tangible that would really help our audience.

Boyang:
Yeah. I think it's, let's go back to August, right? So in August my co-founder and I were Playing around with a whole bunch of ideas.One of them we settled on was this 2D textures for games. So he has a background in, in 3D games and at Facebook running Oculus. And so what we did was we just made a very simple, page with you enter something and you get back a texture, that book like that thing. so that was our first foray into any of these newer diffusion models. And from there we started. Seeing the same sort of, using the same backend, toward different customers. And this was really just week after week of iteration where we strategized and we said, Hey, what is the thing we wanna focus on? Oh we believe that workflows are the thing that really matter, right? You have to get embedded into your customer's workflow. So if you make something where they come and they get their thing and they just like immediately leave you can get a lot of people coming to your site, but it's not it's not gonna grow into anything. Substantial, All right. So we targeted the workflow we started targeting the customer, which we said was e-commerce mostly for the large market. And then the actual product stack itself. We sort of device an engine to create apps very easily where it's the top level is UI that we build with Chakra. So it's very fast for us to just iterate on this thing.

Dhaval: What was that Chakra?

Boyang:
Chakra is a, like a UI toolkit. For React. I think it probably supports other frameworks now. And then on the backend for us, uh, we have essentially a, a lot of APIs that go to different, ML models that we've stood up ourselves. And so as we keep building, different services, we stand up more of these models. And then we just have access to so much flexibility at this point where we're like, Hey, we need to upscale this image. But we already have that, right? We built this for another reason entirely, but as long as we can leverage the same flows, um, that will just work. Yeah, I think if that answered the question.

Dhaval:
Yeah, that was very helpful, Boyne. Thank you. So there has been this debate about building on top of existing ML models versus standing up your own ML models and your own ML infrastructure, which route? Should a product creator take if she wants to build an AI product, what is the best, what is the best way to test your hypothesis? And at the same time, what is the best way to actually create a high efficacy product? If you were to offer some input there?

Boyang:
I think it depends on where you think your talents and expertise are. For us, we're both technical. It was pretty easy to set up these models on our own side. We are not training them ourselves, right?And so , we're using open source packages, but it's easy to set that up if you believe that your expertise is in go-to market and, SEO and you just want to set up things that provide value very quickly. But you don't care about really optimizing the underlying stuff as long as you can get the users. Then Sure. I think that route is really good. I would just say the most important thing is for us to understand what our strategy is. And the strategy has always been, hey, like we are two technical people and we're focusing on distribution because that's sort of our weakness at this point. So we're trying to short that up. And that's why we made decisions we did. But if you are, coming from product and you just wanna stand something up these days the models are so good and there's so much out there in terms of SaaS that just lets you, get value immediately and get APIs immediately. That I think is the better route if you were to want to iterate really fast.

Dhaval:
Very cool. So either distribution could be your advantage, or the core ML could be your advantage. You gotta figure out which path you want to take. It could be both, but you have to choose one at a time and then go forward and then iterate from there. Did I get that right?

Boyang:
Absolutely. I think for us we very early on we said, Hey, if we're going into this ML space neither of us is a PhD, right? We're not truly confident in our ability to produce net new models or like net new ways of doing things, however, We are very confident in generating better UX in like allowing people, and this is the whole ChatGPT argument, right? Like it's better UX on top of something that is roughly the same and didn't really take that much, but it was a step function, difference in value. And so that's been our thing, like we're gonna leverage this ux and build as fast as we can to get the most users. And then the models will come. They'll come later.

Dhaval:
Wonderful. Now, changing gears a little bit, are. Generating revenue as of now. What is the price point? How did you come to that conclusion? Are you pre-revenue, post revenue? Tell us a little bit about your revenue stuff.

Boyang:
Yeah, pre-revenue. We're basically concentrated on gathering users. so we have built up a, an email list, of users. And the point of this is like to get iteration as fast as possible to come to a good pricing model. And we have one in mind. We're building it out. I think we're gearing up for this product hunt launch that will come in like the next week or week and a half. so I think that is the point at which we will find more legitimacy in saying, Hey, we have value to offer, that you can pay for. Right now it's a bit early and it just seems like. If we were to ask for the money, it'd be, we're not really providing that much, so there's not much to pay for. so that's the revenue argument. I think also for us, if distribution is so important we don't really wanna slow it down with anything. we would rather people see us as, Hey, we remove backgrounds really well and we give you like a flat white picture with nice shadows of your product and that's free. And if a million people use that, then we're very happy with where we.

Dhaval:
Wonderful. Tell us a little bit about your venture, uh, funding strategy. Are you, have you raised capital? Are you raising capital? How is it compared to your past startups in the current environment? Yeah.

Boyang:
I think funding We know the funding environment's a bit tough. it might not be as tough as it was, like the end of last year. we are not seeking money right now. I think we want to be in as strong a position as possible while going into that fundraising. And for us it's like we have milestones that we know we want to hit, before we really had conviction that our product is good and, you know, will have product market fit in the future if we throw money at it.. So right now, money's not the problem. we want to get to enough users, enough retention enough growth on the payment side, in order to say, Hey, like we know now the equation is if we go seek fundraising, we get this like $3 million. Here's what we are gonna throw it at and this is how it will help us grow.

Dhaval:
Yeah. So becoming clear on where you want to invest is where you are at right now. So do you have any funding, have you done any seed rounds or, is it all bootstrapped until this point?

Boyang:
so, last year my co-founder and I joined a, I would call it an accelerator. It's a community, sort of like Y Combinator, it's called South Park Commons. been really great for us as like a networking opportunity. and then that is where we got our umbrella. It's like a pre-seed. it's really just a parachute for us to, you know, jump off a cliff. So both of us were coming from relatively big tech jobs, so pretty cushy jobs. And, that was what gave us the confidence to just be like, Hey, let's let's quit. Let's do this instead. Um, and yeah, that, that money has gotten us to where we are. We're not really, we're not paying ourselves very much, so we have plenty of runway. and you know, for us it's really. Like we're not looking for like the big investment dollars. I think it really makes us happy to just build things that people want.

Dhaval:
Wonderful. Tell us about the future vision, like really long-term future vision of, stylized.

Boyang:
Yeah, so we want stylized to be an asset first e-commerce platform. And what this means is I can imagine a future where you wanna start a shop. And you don't go to Shopify because you need a website builder or a domain name or any of those payments options. You come to us because you have the thing in mind that you want to sell. And you can immediately launch the shop of your dreams, right? That is online. And so all of that virtual stuff, I actually don't think, and a lot of shop owners have told us this. The boiler plate of setting up what your shop, looks like necessarily on, in virtually is part of what they want to do. They're passionate about making good product, getting customers selling that product and like iterating on that. And that's what makes them happy. And we want to take away like, All of the abstractions up until that point, right? So we want you to say, Hey, I have a good product, or even I have an idea for a good product. What should my pricing model be? How do you do the SEO for this? What should it look like? How do I set up a website that is visually enticing to anyone who comes in? And once we have, I think these better image models, we'll be able to even A/B test the actual images themselves, which is something I think up till this point no one was able to do.You can't just run to a photo studio every week and be like, Hey, give me like five sets of different photos I want to try out on different users. But what were you willing on is like, Hey, this seller likes these photos. We'll just, on demand, we'll generate those of your product, and present it to them. And I think that is the real e-commerce experience that we're going for.

Dhaval:
Wow. Okay, so you're starting with The seller's mindset. The seller's product mindset first, which is, I have an idea of selling this product. Help me launch a store, get a website, get the product page up, get the content up, and I can just focus on the product innovation. Did I get that right? Are you focusing on like taking over? All the abstractions, including the e-commerce, including the website, including all the marketing, including all the content. Did I get that right? Is that your vision?

Boyang:
Yeah, that's the vision. And I think the way that it's been working so far, in the last 20 years is we've been able to abstract away some of these things. As a startup founder, I didn't have to set up my own HR ops or any I used gusto, right? And so these layers exist now and I think they don't quite exist from the retail to e-commerce side. You can't, you can see that transformation just about to happen and we want to be, To really transform e-commerce into something where you can say, well, like I wanna stand up an e-commerce store, and I click a couple of buttons and I tell you my idea and now I have it.

Dhaval:
Wow. I'm excited about where you're going Boyang. And looking forward to your launch here. Good luck with your product launch. We'll be keeping an eye on awesome things you're working on and thank you so much for being part of this show. It's an honor to have you on the show.

Boyang: Yeah, absolutely. It was great to be here.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
How founder this build a website copy and image generation plugin
31 mars 2023· AI Product Creators
Andrew Palmer is the Co-Founder and CEO at Bertha Ai. Single Dad of a Twenties daughter He love to travel, play golf and attend WordCamps around the world. He love supporting Plugin and theme developers across the globe as it helps them get a profile and earn a living from their knowledge of WordPress, AI Content, PHP, Laravel and JavaScript. In today’s episode, We discusses his journey and the development of Bertha AI, an application layer built on top of OpenAI's GPT-3. Bertha AI is designed to help website owners create and manage content efficiently. Andrew shares his advice for new entrepreneurs interested in creating AI products to start by fine-tuning their prompts and understanding what they want to achieve with AI. Tune in to hear Andrew's insights and experiences in building Bertha Ai and how you can apply these lessons to your own business.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Andrew:

• LinkedIn: https://www.linkedin.com/in/andrewpalmer/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
[00:00] Introduction to Andrew Palmer and Bertha AI
[00:04:20] How Bertha AI uses GPT-3 for content creation
[00:07:47] Andrew's entrepreneurial journey
[00:13:58] Advice for those looking to create AI products using GPT-3
[00:17:17] Staying motivated as an entrepreneur
[00:19:31] The future of Bertha AI and content industry

Transcript:-

Dhaval:
Hey, Andrew, thank you for joining the call. I would love to hear from you on what your product is, like who does it serve, and what are the things that it solves for?

Andrew:
Well, Bertha.Ai solves a multitude of problems. In WordPress, for instance, we have web developers dealing with clients that really just can't get write together own content because they're not concentrating on what they should say on a website. They're concentrating on what they should say to their direct clients. They're speaking to every day on the phone or in day-to-day meetings, sales meetings, networking, that type of thing. So WordPress has a content gathering issue with Bertha.ai, when you are developing a website, the developer and all the client can use Bertha.ai to create interesting content around what the services, the particular website client provides. It can also help them write blog posts relevant to their particular industry. It can basically help you get the right words out there at the right time and much, much faster. So you're going to increase your productivity as a web developer and also if you're a client building your own website, you are going to be able to increase your productivity and get more ideas about what people actually want you to talk about within your services. So that's what Bertha.ai is all about really.

Dhaval:
Wonderful. Is that a content management system or is that a plugin for Wordpress?

Andrew:
So Bertha.ai is a plugin for WordPress, which is an application layer built on OpenAI, which everybody's hearing about the moment, probably one of the best viral campaigns out there with GPTchat. With Bertha.ai, we've got a number of modules in there that help you get your unique proposition together. You can write a full on about page. As of this week, which is January, 11th of January, so coming soon is GPTChat to Bertha.ai as well. And as of by the end of January, we'll have a Google extension out there as well. So you'll be able to use Bertha.ai everywhere. And the whole point of being able to use it everywhere is that not every website is built using WordPress. So you'll be able to use it in Shopify, Wix, those kind of proprietary website building tools.

Dhaval:
Wonderful. So you just answered my question that you are expanding for being a word, from being a WordPress only product to more of a open access for all types of content management systems.

Andrew:
Yeah, using it everywhere. So when, if you're in your Shopify site, you'll be able to just invoke Bertha .ai and write fantastic product descriptions or enhance your product descriptions. You'll also be able to get page content that is relevant to your users. So you'll, you'll ask one line question and Bertha.ai will be able to give you the answer to that question. And then you can copy and paste that, edit that to make it more human-like if you like. And then you can post that in your Shopify website or your Wix website or Squarespace, whatever you're using.

Dhaval:
That's wonderful idea.Tell us a little bit about where you are in your journey. When did you start first? How many customers you have? If you can share revenue. What is the split between developers versus end clients? From a business point of view, tell us or give us a few clarity on where you are in your journey.

Andrew:
Where we are is we're nearly a year down the line or just over a year down the line. Actually, we launched in September 2021. So, just the MVP version, and we launched that for basic WordPress plugin. So we're a year down the line. So we're way down the line. We've made the plugin faster, more accessible, easier to install. Getting the customer journey right is very important when you are building a plugin that's got a lot of things in it. There's also a learning curve as well. So we produced in just under a month, we produced 54 learning videos which are on our YouTube channel, which take you through every single module. With the hype around GPT-3, obviously lots and lots of people, you've got a million users in under three weeks, I think on GPT-3. So people are understanding how to use AI, how to ask the right questions, and obviously with images, image generation as well. So Bertha AI put image generation in there about a month ago. And that's flying. People are really kind of intrigued about how they can create unique images for their blog posts, from CVD products to computer products to any kind of product for florist, interior design, kitchen design. You can really get some great imagery there. If you want a modern kitchen design designer or a modern interior design, just ask Bertha create an image and it'll produce that image of a beautiful laid out sitting room or a house with chandeliers or whatever you like. So the point is that we came from the WordPress design or website design business. So we understand how people want to build websites. It's kind of a generic way to do that. And with the developments that we've had, we've got something like around 10,000 registered users. Some of them are website owners. In fact, I had a Zoom call with a guy today who's a website owner. They're not a developer or anything. And he wanted to use Bertha AI to better describe his products to his wider market. In fact, most of the meetings I've had with Bertha AI I love doing one on ones with our clients. And they've been with people that aren't web developers. They're the website owners. And that's the people that we really like to target. Because as I said, website owners have a real problem getting content together and writing blog posts. So Bertha helps you with all that content. But revenue terms, we're okay. We're fine. It's in profit. It's running itself. It's quite nice, we're able to expand. Development team based in UK is upon part of the development team as well. And the majority of the team are based in Kolkata in India. And they do a great job.

Dhaval:
So then quick question. You mentioned something very interesting. There are a couple things you mentioned. I want to glean them out. One is the customer journey. The audience for this podcast are people who are product creators and they're specifically interested in either adding AI to their existing product or they want to create an AI product. With what you just described about customer journey, what was that experience like for you? Like, understand the customer journey for your website owners. And how did you productize that? If you can share a little bit about that.

Andrew:
Well, it's still happening. with the WordPress plugin, we are lucky it's on the repository and you can install that directly from your WordPress dashboard. With the extension, that's going to be a harder task making sure that customer journey is OK to actually install it in a Chrome extension. We're talking about millions of users that maybe don't know how to install a crime extension from scratch because it's not going to be on the Chrome extension store for a while. It's just going to be downloaded. Once you have an account within Bertha, you'll be able to download the Chrome extension and install that. So we're building education around that about how to install a Chrome extension and maybe make it not so difficult for people to understand that it's quite easy. It takes about 30 seconds to install the Bertha extension. So it's about education, but it's also about what we want as a customer journey. There are so many extensions out there. There's so many plugins. There's so many SaaS products out there that involve a learning curve. And what we've learned is that people don't actually want to learn. So what we have to do is almost automate that process .

Dhaval: I'm one of them.

Andrew:
Yeah, exactly. You don"t actually want to learn how to drive a car or drive a new car, let's say. So if you've got a new car and it's great, but the indicator arm is on the other side, you forget, don't you? You put your windscreen wipers on. So, There's a learning curve around everything if I really simplify it. But at the end of the day, it's up to us as product developers to make sure the customer journey is as seamless as possible to immediate use. And that's where we're having difficulty. And we've had difficulty, but we're getting better and better by that every single day.

Dhaval:
Now, you were a website agency website development or marketing agency, and that's how you identified this content problem that your customers were having. And that's how you ended up with this idea . Am I getting your entrepreneurial journey right?

Andrew: I mean it was my co-founder Vito Peleg who has a collaboration tool with wordPress as well, and with websites now is a whole, where people can leave feedback actually on the website. And he said, well, wouldn't it be nice to have some AI generate text automatically so the customer doesn't have to worry about what text to put in a particular place in a website? So we developed it for that purpose initially and then thought, well, you know what? We can make this work with every single page builder out there. Gutenberg, the classic editor for WordPress. And that was a learning curve as well. I come from a thing called Divi theme community, which is by elegant themes. So my first task was to make it work with Divi theme. And I've also got something like 16 plugins out there that work with the Divi theme. So we know that theme pretty well and how to make things work on the front end. And the nice thing about using page builders, most of them are front end. The user interface is front end. So what we've done is we've made Bertha integrate with every single page builder that people use so that when they're building their sections together, they can actually put the text in the same time as they're building their sections. So that was the key to us to make it user-friendly. Page builder friendly. Certainly Gutenberg friendly with WordPress, because the way that Gutenberg is progressing and systems out there like the kadence system for WordPress. It makes it easy for people to use blocks now because kadence is really the lion of where blocks are going. So we had to make sure that it works with every single theme out there, every single page builder as well. And there's some page builders that have just come out recently that we're still working on, but it works. One of the most helpful people out there was a company called Visual Composer who helped us integrate Bertha into their system. And that then led us to realize what we were doing wrong with other systems that were built on React and other technologies other than just the normal page builders. So that helped us a lot. And, Yoast.com as well. They helped us integrate it into the Yoast plugin for SEO as well. So there's some good teamwork out there. And I think that's what product makers need to realizes is that the way that they can really help their products progress is to get involved in partnerships or relationships with successful products out there and make sure that you keep that conversation going. That then leads to them once your product works seamlessly with their product that makes them help you promote it as well. So that's a good tip.

Dhaval:
Wonderful You just gave a golden nugget here. One thing I eased out from this is that Most of the people who want to start out as a product creator, they end up choosing something that they already worked in. For instance, in your case, you chose something that you already had experienced in and you found gaps and you were able to solve that problem by creating partnerships with other people. So it's a lesson for the audience, if you are starting out, look closer, look close at your home. Where are you? Like where in your work, in your current work. There is an opportunity where you can use AI. And that's the starting point, like you did. Now big question for you is it seems like you were able to successfully bootstrap this and you have revenue and you have users, 10,000 users. That's a success story. Did I get that right? Or did you raise capital or did you use your agency money to build a product? Help us understand, how did you financially drive this forward?

Andrew:
We've bootstrapped from the very beginning. One of the things that's prevalent in our industry or the plugin industry is to do a lifetime deal. So once we had the product at a stage where we knew that it was a flyer or we knew that it was gonna go along, we did a couple of lifetime deals. So that was kind of self-funding. Our lifetime members really invested in us as well. If you think about it, it was a good lifetime. It's a great lifetime deal. And we're always trying to improve it. We don't do lifetime deals now. So that gave us traction.

Dhaval:
Now where this Your agency clients who signed up for your lifetime deal or did you go out and did some marketing campaigns?

Andrew:
We did some marketing campaigns and just said, This is where we're at. We're pretty well known in the Facebook groups where I get on podcasts. So for instance, the time now is three o'clock in the morning. I don't mind when I'm going to talk to people on a podcast because I'm based in the UK. Bertha is an American company. We registered in the US. But the point is that you've got to get out there and say, let me talk about this product. Let me talk about the benefits of using AI, even though there are some negatives to it because of People are saying there's repetitive text. But again, once you educate yourself as a human, how to use generative text within AI, then you're going to produce great content and you're going to be a leader in getting really informing, educating, and entertaining your customers. And that's the whole point about generative AI. But what we wanted to do was totally bootstrap. I was very lucky. I had a business called Elegantmarketplace.com. I sold that to in motion host and that gave me some financial backing as well. So I didn't really have to worry about income for a little while. So that meant that I could concentrate with my team and I on developing Bertha. But I've still got my web agency in in the UK and I service around 500 clients on hosting and maintaining websites. And we pick up the odd website here and there. But Bertha AI is really pretty much my full-time job these days.

Dhaval:
What would you recommend for someone who is starting out and has interest in creating a product, an AI product on top of GPT-3 or any of the future versions? Would you say that's like a solid strategy to put a user interface on top of GPT-3 or would you recommend building your own models and having that own core machine learning? What is the trade off there? If you can speak to that please.

Andrew:
Well, the trade off, I mean, is you can do some fine tuning. We've done some fine tuning on our prompts. I mean, the whole point about Bertha AI was. It was developed specifically for website owners. So the prompts come from our life experiences within developing WordPress sites and writing and copywriting with WordPress websites. So we did a fair amount of fine tuning, which can be get expensive. But we also knew what we wanted to get out of Open AI. So we've been playing with it since May 21, and we didn't develop stuff or we didn't launch anything until September 21. So We had a real good session, a few months of learning how to write prompts that worked properly and developing prompts and producing new items like ask Me Anything. We've got a module within Bertha AI. Now, which is effectively chat GPT, where you can ask Bertha anything. You can ask it to write a song. Write a bit of code. The code may not be right, so you should always double check. But write a 500 word article on SEO, for instance, or product development or whatever you may want to do. Now we are adding GPT-3 chats. That's got even better. So we're keeping the asking Me Anything module. We're keeping the long form. We're keeping all the other modules that we've got. And also OpenAI helps you write in currently we're baked in 27 languages within Bertha AI as well. So when you are thinking about producing a product, make sure that you internationalize it. That's one of the keys. Make sure that you've got a global audience rather than a local audience. There are plug-in developers out there product developers out there that really niche down. There are people out there that are selling to enterprise only with their extensions or with their product solutions. But if you want to be able to get out there to the masses, one, cost it properly. Make sure you understand your costs completely. And two, price it correctly so that you understand where your profit margins are going to be so that you can continue to make a difference to people's lives. Really, that's the key. It doesn't matter what you develop, whether it's an AI product or whether it's a theming solution or helping people meditate properly or be more spiritual or less spiritual or whatever you like. You can use your intelligence. Let's go away from AI at the moment. The human intelligence is you just do your own research or what are people using and why are people using it? And if you want to use AI to help you build it, that's fine. Let's not forget the mRNA vaccine wouldn't have been produced in a year without the power of AI. So there's plenty you can do. You can go into healthcare, you can go into well-being, you can go into fitness. AI will help you build a product around anything. Which is great.

Dhaval:
Andrew, I have two last questions and we'll wrap from here. So one is you're obviously not in your early 20s. You're not a single guy or a single gal who has a lot of time and a lot of energy, right. How do you keep your energy and your fire stoked as an entrepreneur?

Andrew:
Okay. I get asked that question a lot. I don't really have an answer, apart from if you love what you do, What you do will love you back. So if you love what you do, what you do will love you back. I am an open source advocate. I'm a community advocate. I love what goes on in the community of certainly around WordPress, in the community of AI. I'm an anti keyboard warrior guy. I can't stand keyboard warriors. My goal in life is to calm people down, basically, make sure that we've got the facilities to be helpful. Bertha AI is essentially, So it's nice to be able to let people know that they can use it free completely forever. There's no restrictions on use currently. So I just try, I just know that what I'm doing in my life and in my business. I know that I'm doing it for profit, but I'm also doing it from an altruistic sense of what's right. And that's been learned over a period of years. It's not easy to get to give away things or get to give away your time. As I say, one of the most rewarding aspects of my as an entrepreneur is actually helping other people. I have a coaching business and many times I look at people that are in trouble with their business or anything, and they've come to me and I charge an hourly rate for that and, You know, I just don't charge. I just go, okay, you can have that one on me. Because, and it's not because I feel sorry from it, it's just because I feel I just want to give them an opportunity to come back. And, conversely, what then happens is when they do become successful, because they do, they come back and they pay me my hourly rate on a weekly basis. So, there's give and take in this . So basically wake up in the morning wanting to give rather than take. And I think that's the best entrepreneur that you can have. Because in the end things will come back to to you.

Dhaval:
That's beautifully you said, I love your quote as well. One last question and we'll wrap. What is the future of Bertha? Where do you see it going?

Andrew:
The future of Bertha is going to continue to be an application layer. We may very away from Open AI. What Open AI have done with GPT chat is almost a kick in the shins to the people who have built application layers on top of Open AI. They gave us a bit of a chance, but if you think about where open AI is going to go, they're going to probably get another massive investment from Microsoft. They're going to get another massive investment from anybody else. So our job as Bertha is to make sure that we exploit in the best possible terms of that word. We exploit what Open AI are doing, what other people are investing within Open AI and make sure that our application layer is as proficient and efficient as possible and brings new things. So one of the new things that we are doing is obviously bringing So I'm not going to go too much into it, but something that will really help the podcast industry, which is a billion dollar industry out there. So we're developing things that will help people really make sure that their podcasts are transcribed properly, got the right SRT fields files, and with video transcription as well. So all that kind of stuff moving down the line. So making Bertha AI from an extension and a plugin, a WordPress plugin. Let's not forget there are millions of WordPress websites out there that need good content or need their content looked at. So we're just going to continue to promote the benefits of Bertha AI to website owners. We're also going to promote it to podcasters, to people like you that are taking your time out to interview people like me and making sure that you benefit as much as you can from using AI and hopefully you benefit from using Bertha AI. We'll keep innovating as much as we can rather than following. That's the key.

Dhaval: That's exciting, Andrew. I can't wait to hear all about what you're working on when you come back again on this podcast a few months from now and hear about all your success. Thank you so much for your time.

Andrew: Thanks For having me
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI product to replace your employer with AI.
30 mars 2023· AI Product Creators
James Clift is the Founder at Durable. He is the former Founder at KarmaHire, WorkStory. VisualCV and Holopod. He building businesses since 2005. In today’s episode, We discusses the benefits of large language models (LLMs) like GPT-3, emphasizing their ability to generate human-like text that can be used for a wide range of applications, including content creation for websites. He also emphasizes the significance of having a data-driven approach to fine-tune the models and make them more effective for different business categories. James shares his experience of finding partners with AI expertise, highlighting the importance of networking and being involved in communities like South Park Commons. He believes in sharing work and attracting like-minded individuals to join the venture. Tune in to hear James's insights and experiences in building Durable and how you can apply these lessons to your own business.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find James Clift:

• LinkedIn: https://www.linkedin.com/in/jclift/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introduction
00:03:50 - The story behind Durable and its AI-powered website building platform
00:07:22 - The importance of user feedback in AI model training
00:10:49 - Fine-tuning and prompt engineering for LLMs
00:11:22 - Best practices for partnering with technical teams
00:12:27 - Durable's growth, funding, and team size
00:13:20 - Remote work culture at Durable
00:13:49 - Finding AI expertise and attracting technical partners
00:15:16 - The value of sharing your work and attracting like-minded people
00:16:20 - The future vision for Durable

Transcript:-

Dhaval:
This founder built an AI product to replaces your employer with ai. Yes. You heard that right? Your employer, James, is our guest in today's show, he shares his learning on how he built a product in a hyper-competitive market space. James Clift is a founder of durable. Durable, makes owning a business easier than having a job. He's a former founder of KarmaHire WorkStory, VisualCV and Holopod. He has been building businesses since 2005.

Welcome to the call. James, tell us about your product.

James Clift:
Awesome. Yeah, so Durable is the fastest way to build a website on the internet. In three clicks, in 30 seconds, you can generate a business website using ai. And not only can you make a website, we've got the rest of the stack as well to operate a business, we've got a CRM, an invoicing tool a financial account. So essentially everything you need to start and grow your business in just a few click. .

Dhaval:
Wow, that's very powerful vision where do you sit in your market space? are you serving a specific customer segment?

James Clift:
Yeah, so we're totally focused on solo operators, so anyone that runs a solo business. So primarily those are service-based companies, so everything from a web designer, marketing contractor, copywriter to more traditional physical service companies like Lawn care, home Services. Plumbing contractors, skilled trades. So essentially anything where you're trading your hours for dollars or your hours for projects we're a great fit for if you're selling goods on the internet or you have a brick and mortar store, we're not the best solution there, but solo service-based companies is our primary market right now.

Dhaval:
How did you differentiate yourself in this crowded market space? I believe it may be high competition market space. But You have, you seem to have found reasonable amount of success based on what I have seen about you online. How did you create that differentiation for your product?

James Clift:
Yeah, I think there's a lot of ways to look at markets like most markets are very large on the internet, and for us it was a few things. So one is bundling, so providing all the tools you need to run your business under one login. So that was a big value add from the start. So you don't have to learn five different tools. You don't have to pay for five different subscriptions or 10 different subscriptions. It's everything you need under one platform. And then the other piece is what can you actually do 10 times better than everybody else? And for us it's the speed of actually getting a website. Out to market. So instead of taking, typically it's weeks to get a website live, if you're really good it's days. If you're really, really good, it's hours. We actually do that in minutes. So there's this order of magnitude that makes that thing faster, or that business process faster. And it actually unlocks a lot of creativity, a lot of, just makes it more fun and playful, not stressful. And I think you open up these brand new markets by just anytime there's an order of magnitude step change. Something's 10 times faster, 10 times better, 10 times more intelligent. That creates these huge opportunities. And I think AI as a platform is definitely one of them that we're seeing. yeah, I think customers are super excited about the speed, the simplicity, and then the bundling aspect of the platform as well.

Dhaval:
Very interesting. So you brought this up, AI and this particular discussion and discussions like. That I host are focused on people who are either interested in creating an AI product or infusing AI in their existing product. Tell us about your infusion of AI into your product. When did you decide that, was it AI first from the ground up? And if it wasn't, when did you decide to bring AI into the user?

James Clift:
Yeah, I think, so I've ran SaaS companies for a while now, so probably about 15 years. And, it's always, I mean, the goal of any software company is how do you make processes easier and make your products easier to use? So, the long-term vision of us and AI is really, How do we replace your employer with AI and just let you focus on your core competency. So that's your skill, right? So a lot of the time if you have a service job, you have an hourly rate that you're then getting marked up for by your employer. So, in a perfect world, you just meet that reach, that market demand. So, hey, you're a lawyer making, I don't know, call it 500 bucks an hour that your employer charges you out. You're making 200 bucks an hour. So that's a market opportunity for you to go independent and do your own thing. But what the law firm brings to you is a brand customers, some back office services. So the way we're thinking about that is what can we actually replace? And I think brand is changing a lot. Like the brand matters less, the individual matters more, and the back office piece can be solved with technology. So essentially . Everything can be automated except for the thing you're really good at, and that's really how we're thinking about ai. So from a product standpoint, we built the platform first, and then we built the AI on top of the platform because we've got a lot of features and it was always this idea of, how do you make those features easier to use, more accessible, more interesting, and just more intelligent. So the, our customers can just focus on what they're good at. so that's always been part of the strategy. Definitely it's accelerated in the last few months here with all these new technologies and APIs and libraries that have come out, and been super incredible and are moving really quickly. So definitely, the primary part of the strategy moving forward as well.

Dhaval:
One thing, one thing that I always hear from other product creators in the space is about finding the balance between building on top of existing AI capabilities. That other companies have created versus building your own AI capabilities? Where do you draw that line in your product?

James Clift:
Yeah, I think it really depends on what kind of company you want to be. Are you really good at marketing? Can you repackage these libraries and build a good user experiences around them? And you can accelerate really quickly? Are you deep technologists? in that case then you should build the underlying infrastructure layer. For us, I think the advantage, and I think. These core models are really, really powerful, but I think you have to train them on your own data sets. Otherwise you're gonna lose your competitive advantage really quickly. So if you're just re-skinning chatGPT and it's a slightly different user experience, but the same data, That's gonna be a race to the bottom pretty quickly, because everyone can do that. But if you have a user that is unique that you can build data sets around, then you can train these models to be more effective. So we're doing both, we're using the existing models, but we're also training them. Pretty specifically around our category of customers. If you think about a solo business owner, there's a set of activities that you need to do. Even from a marketing standpoint. It's okay, you've got your website. How do you optimize your seo? How do you create your ads? How do you create your marketing copy, your newsletters? Once you have customers in, like, where do you get more customers from? How do you measure your channels that are effective? When you send invoices, what is the value of that invoice? How does that tie to your accounting system and your customer database? So there's just a lot of things that we can actually build more proprietary, unique data sets around and workflows and processes that we can optimize. So I think as long as you own that customer journey and lifecycle, then you have the ability to train your model and make it better. But yeah, if you're just re-skinning an API, I think. There's some that will do well. So like re-skin the API in the category. I think one or two of those will succeed in every category because it is great tech. And if you're good at marketing and acquiring customers, there's opportunity there. But if you're the, I don't know, the 20th company to try and build a copywriting app with ai, I think that's gonna be more challenging.

Dhaval:
Thank you. Thank you for sharing that within your customer journey. Where do you bring ai? Is it from the get-go or is it, how do you find that balance to delight the customers versus, make them anxious?

James Clift:
Yeah, right from the start, I think, It's just a really cool technology to see, and from what our customers see when they're getting their websites built by us, you can actually see the copy being written and it's always wow, this is surprisingly good and really cool. So I think it's just like exposing them to the magic of this technology as early as possible, but it's not, but not trying to make it inaccessible. Right. I think the beauty of ai. Is that you see the output, right? It's not scary because you can see, oh, this is just copy. These are images you chose. This is a marketing strategy that you came up with. It's more just impressive than scary. and I think, yeah, for us it's really giving them that front and center and the first part of the user experience that's really delightful and giving them something to work from that then the human, I think humans are a lot better at editing than creating. A lot of the time. So give them something to play around with, to edit, to regenerate, just to make it fun as opposed to this tedious thing that you have to learn. It's more of a playful experience, I think is what we're really trying to build for our customers.

Very insightful:
So making it playful is one of the ways to pull customers in so that they, and also making it transparent so that they see it happening in front of their eyes and it's a magical experience rather than. Behind the scenes, here's what we are doing, take it or lose it, or any of that. One thing, one thing often comes up is the thing that you brought up is fine tuning on top of existing LLMs. Do you have any tools for other, do you recommend any tools for other AI product creators in this space who want to embark on that journey? What are some of the quick ways that they can fine tune on top of existing LLMs?

James Clift:
Yeah, I think that's the challenge now is that space is so new that, tools are popping up every day, but nothing's really. Taken over as the category leader yet. So I'd say just the more you actually, I mean, prompt engineering is a career now, so being able to figure out how to write good prompts for open ai, I think is a really interesting way of thinking about it. We're building a lot of that in-house as well. Like our own like model optimization is all gonna be in house at this point. I think when there's third party services, we'll definitely look at them as well. But I would say just use. Underlying technology. and then under, the more you understand it, I think the more you can optimize it as well. So play around with it, test it out. If you wanna read the papers, they're actually really interesting as well. I think from a. Like a pure software product. There's not a lot that I've seen yet. I think it's all coming out and people are building it. There's prompt analytics and stuff as well. That's really interesting. But for now, I think, yeah, that's where the expertise comes in, right? Is how deep can you go and how deep can your team go on these different models and training and all that.

Dhaval:
Now as a founder, are you heavily involved in the product at this point or have you gotten to a point where you're focusing more out, focusing out? Or are you focusing in on the product and you have a separate product manager leading the product?

James Clift:
It's, myself and my team leading the product. So we don't have any product managers or project managers right now. So yeah, we're still really heavily involved in all product decisions. We've got an incredible UX and product designer, incredible engineers, incredible marketing folks. So I think it's more, it's an, all the whole team's working on it right now. I think it's led by, the group right now and at the end of the day, it's me that's making the final decisions on product here.

Dhaval:
Any learning lessons you wanna share around fine tuning and prompt engineering when it comes to building on top of LLMs. How has that experience been with your engineering team and what do you recommend others to, what are some of the practices you recommend other PMs or founders to model if they wanna establish that strong partnership, with their Technical partners, to fine tune and create a strong AI product that creates differentiation. What are some of the best practices for that partnership? What does that look like?

James Clift:
Yeah. I think it's really just having this culture of experimentation and iteration and testing. It's really just volume. How much can you test different models and outputs and questions and, gathering feedback from users as well. So having a mechanism to gather feedback around those prompts. And then having your data. So if you say, Hey, we're. We're doing this right now as an exercise. we've got thousands of categories of business being created and we're saying, okay, we're really good at creating lawn care websites, but we're not so good at personal trainer websites. Why is that? Maybe the images we're choosing are off, or the prompts we're using don't apply as much. So it's really just training your own data set and having that data available and also the simple answers look at the output and decide if it's good as either get the feedback from the user, get the feedback from yourself. But I think being, pretty honest about like, okay, this is actually good or not. If not, what can we change and I think that's what we're finding, with some of our categories. Like, wow, this is totally off. And then you really dig in and evaluate what went wrong there, what you can fix. But really it's just volume iteration. And just, yeah, playing around with the outputs and the options. It's pretty much it. Put in the work basically. No simple answer. ,

Dhaval:
tell us a little bit about the size of your team and a little bit about where you are in terms of product. Have you raised capital? Are you bootstrapping? Are you looking to raise capital? How many users you have? What is your revenue like, and what is the size of your team.

James Clift:
Yeah, so we raised capital in May. We're eight people now. Raised a 6.2 million seed round from some awesome investors. We are at about 200,000 websites built, in the last three months. So we launched in October, and that's just been doubling month over month or tripling. So yeah, lot of users, lot of, a lot of, we got some revenue coming in the door now we're not disclosing that. But it's growing very rapidly as well. So yeah, all systems go right now. And the plan is to grow the team right now as well. I've got a few new hires coming on board pretty quickly, so lean team right now. But yeah, definitely operating very efficiently and shipping product very fast. Right now,

Dhaval:
are you a hundred percent remote hybrid? Where are you hiring?

James Clift:
Yeah, we're remote by default. So we've got, team members in South America, Vancouver, San Francisco. And then we do have small offices in those locations as well. So kind of your hub so you can come in, do work, or work from home as well. So we design our whole culture to be remote first, so a lot of async, not too many meetings, but some crossover time for collaboration. and then, yeah, we've got a space you can go into and work, but it's not required it's optional

Dhaval:
as an AI product creator, did you have that AI domain knowledge or did you find partners who could fill in your entrepreneurial gaps for that area?

James Clift:
Yeah, it's definitely, it's always been an interest, but not an expertise. So it's something that I've like it's been a hobby, but not part of any business that I've done before. So it's really just, yeah, finding the partners and the technologists in the community that, is on the cutting edge of this stuff. And yeah, there's a few out there that are great. and I think like one, one of my early investors is called South Park Commons. So they're a community of pre, before it's a zero to one stage. So before you actually start a business, it's a place to experiment with different ideas. And that's a really strong AI community that we've really drawn on for hires and expertise and just learning as well. So I think having a community to learn with is super valuable. and just, yeah, again, being curious and being obsessed with it to some degree, always comes in handy. And I think if you're gonna bet on a technology, it's a pretty interesting one for the next big cycle of software and technology and yeah. What the world might look like.

Dhaval:
What was that process like for you to bring someone with an AI expertise on board? Did you meet them at South Park Commons, by the way? Great Community. South Park Commons is the. Where I have gotten a lot of guests from, and yeah, great community. So is that, would you say you, you attracted your technical partners there or what does that look like? How did you pull them into your vision?

James Clift:
Yeah, I think it's just, again, volume, like having as many conversations as possible. My CTO I knew from my last business as well, so I worked with him before. And he's got his own network as well, so I think when you get one. Really excellent person. They typically bring more excellent people. So that's the value in hiring with and partnering with amazing people. And then just, yeah, getting the, doing all the stuff that seems not that important. So going to the demo, like the little demo days, like getting your name out there, just getting as much exposure on to what you're doing. people who are interested will reach out as well and want to communicate.So not building in a vacuum, but really. Really getting your name out there, putting yourself out there, showing your work. So even if it's not totally done, just ship something cool. And that leads to a lot more than you'd expect. It's the, you can't predict it, but typically the more you're out there, the more people will discover you and the more conversations you have, the better people that you will work with that will buy into what you're doing.So it's not trying to convince someone of what you're doing, but. have a good thing that you're doing and talk about it, and then people will be convinced. The right people will be convinced anyway.

Dhaval:
Yeah. There is, I think there is this famous quote from novel, which is if you share what you're working on, you'll automatically attract people who are interested in that.

James Clift: Exactly.

Dhaval:
Very interesting. Very interesting. What I wanted to ask you next is the future vision of your product. James, where do you see, where do you see your product going?

James Clift:
Yeah, I think I touched on earlier, but essentially, can we replace your employer with ai? So how can we abstract and automate every part of running a business except your core skill? I would love to see. Someone who has a skill sign up for durable and they have a job scheduled in their calendar tomorrow, they show up, they do the work, their invoice is automatically sent and paid for their accounting's automatically done. So you have this total control over your schedule and flexibility. We tell you how to price your service how can you make a business as easy as renting your house out on Airbnb? So really just leaning in on the automation and the AI and just letting someone focus on their core skill. And there's so many problems to solve there. So we're starting with the, the marketing piece. So even from a, like on a website builder standpoint, there's a hundred features we wanna build to make that process much better, much easier. And then once you have your website, how do we help you market? So automating your advertising, your customer outreach, your marketplace connections, your lead generation, your follow ups, then it goes into scheduling. So we're kind of working our way down, the operations stack of a solo business. But yeah, the end goal I think is really, and what we're seeing in the world is hundreds of millions more people that are, that are their own boss, right? They're running their own business. They're more in control of their lives, they're pricing, their freedom, their flexibility, and they're living happier lives because of technology. And that's really where, what we wanna build for.

Dhaval:
Thank you so much for sharing this vision, James. I wish you all the best in achieving that vision. I've done service businesses and I know how difficult it is. You are doing some amazing work there to help individual service business owners achieve their full potential, right? That's where the innovation comes from. So thank you for taking on this vision and thank you for coming onto this podcast and helping us learn from you. Thank you. Yeah,

James Clift: Thanks. Awesome. Appreciate it take care.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI Writing Product that serves 4 million customers and got acquired in 2 years.
29 mars 2023· AI Product Creators
Abhi Godara is the Founder & CEO Rytr. He is also the Founder & CEO at HelpTap. Rytr is an AI writing assistant that helps you create high-quality content, in just a few seconds, at a fraction of the cost! In today’s episode, We discusses the initial stages of his startup, where they utilized organic channels like LinkedIn, Facebook, and Reddit for marketing. He also discusses acquiring training data and recommends strategies depending on the domain, mentioning that GPT can work with a limited number of examples. Abhi highlights the importance of user experience in differentiating his product from competitors. Tune in to hear Abhi's insights and experiences in building Latitude and how you can apply these lessons to your own business.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Abhi Godara:

• LinkedIn: https://www.linkedin.com/in/abhimanyugodara/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introduction
00:03:25 - Abhi's motivation for building an AI copywriting tool
00:05:15 - Strategies for acquiring the first thousand customers
00:09:40 - Differentiating the product in a competitive market
00:12:21 - Acquiring and using training data for AI models
00:13:50 - The story behind the acquisition by Copysmith
00:15:08 - The future of AI in content creation and advice for AI creators

Transcript:-

Dhaval:
This founder built an AI writing product that serves 4 million customers, and it got acquired in two years from Founding Date in this episode, we discuss his product development approach that differentiates his Gen AI writing product from the plethora of other gen AI writing products in the market space. We discuss his product differentiation strategy, his training, data gathering approach, and how he got his company acquired. Today my guest is Abhi Godara. He's the founder and CEO of Rytr and AI writing assistant that helps you create high quality content in just a few seconds at a fraction of the cost.

Welcome to the show, Abhi tell us about your product. Where are you at with it? what's the four 11?

Abhi:
Right thanks Dhaval for having me. so I'm founder and CEO of Rytr one of the largest and probably the first one in the market AI writing platform. We have been there since last couple of years now. now we are serving close to 4 million customers all over the world with with close to perfect ratings pretty much on all the platforms. So it's been an amazing journey in terms of how, the platform has scaled which allows a lot of these content creators. Marketers And professionals to create really high quality copies across a range of use cases, purely through ai. So things like email writing, blog writing product description ads, you name it. Everything can be generated through our platform.

Dhaval:
When did you found the company?

Abhi:
So this was back in 2021 actually when we started working on this. although I've been in the AI space for a long time. but this idea took off only when OpenAI came to life back in 2020. So I was following that closely. And then when GPT 2 and then GPT 3 came out, and we bounced on that seemed like a great opportunity to build something like this and just to give you some background to that. Again I've been an entrepreneur for most of my career. And, when, one thing I've always found that content creation is a pain, especially when you're a small team just starting it's a fact that many startups and professionals fail because they do not possess the effective marketing and copywriting skills. While dabbling with GPT 3 on another sort of chat bot project, I realized the potential of this technology and the market it could address. And at that time we looked around and evaluated existing platforms and found the experience a bit frustrating. And decided, okay, let's give the market what it deserve. And that's how the AI writing tool was born. I think we were probably in the first six months of this technology when it came out. We launched this and yeah there is no looking back since then. From zero to almost 5 million customers now.

Dhaval:
Wow. 5 million customers in less than two years. Did you bootstrap this? Was this venture funded? Tell us a little bit about the financial side of the business, if you may.

Abhi:
Yes, absolutely. So the funny story is , it was completely bootstrap zero external financing or capital reached. We had a acquisition as well, last year now part of a bigger umbrella company called copysmith. And yeah, it was always a small team and even. As of today, we are just four people. It's a very, very small lean team. And for the first six to, I think nine months, it was just two of us, me and my co-founder, and we were just doing pretty much everything. So yeah, it's been a lean journey completely bootstrapped and even as of today we are a very small team that is focused on product and high quality customer support.

Dhaval:
Okay. We'll switch to gears a little bit on. Where did the AI kick in for your customer experience? Customer journey? How did you make that decision that in this point of customer journey will be infusing ai? What was that decision making process like?

Abhi:
Yeah, so I think the whole product itself was like, The foundation was ai, right? When GPT 3 came out, like it, it allowed people to create all kind of content and copies by just giving some examples or you can see, training data so when I played around with the technology, I could see the potential. Wow. What if I can turn it into a delightful experience for end users who can create all kinds of copies. So we did a lot of our own training data in terms of the different kind of copies that people would like to generate we trained the, the underlying sort of models which were provided to us by GPT 3 OpenAI. And then yeah, so the whole product was basically built on that technology from day one. AI was always there. It is an AI writing assistant, it's natural that AI is there. So yeah, so it was always AI first product, AI first pocket you can say. And when we launched, this was just heating up, this space was like just coming to life, I think now. AI and GPT 3 chatGPT all over the news, but maybe a couple of years back it was just a very sort of em embroiling, technology. Not many people knew about it. So yeah, but we decided, well, something like this can really make a difference. So that's how I think we bounced on it.

Dhaval:
Yeah. You mentioned something about you fine tuned the models that you got from open AI. For new and aspiring product creators who may or may not have deep expertise in ai, is that a preferred route? Is it easy to fine tune existing foundational large language models that you get from OpenAI? If there are any tools you could, you would share with?

Abhi:
I mean it to be honest with you, yeah, I think it's, uh, they've made it very easy. So it's not even a non-technical person can feed in some examples and have the AI. Produce content, which is of high quality and aligned with what the user is expecting. so it's not a highly technical of course you can fine tune to the extent that you can provide like thousands of examples with your own custom domain or maybe industry. And then the model would be like very, very customized to your needs. But , we didn't go that far and I don't think majority of the use cases need to go that far unless you're working with enterprises, I guess. so in our case it was, and this was like a couple of years back, now it is matured even further. So you can actually go in with chatGPT or any other such similar technology and with the zero short learning they're able to give you the output that, it's pretty decent. So, yeah. So I think even non-technical founders they're looking to get into the space. I think with some fair like industry experience, they should be able to train the underlying model, which doesn't require any sort of technical expertise. But if they're working with, I think, bigger clients and companies and enterprises, I think that's where maybe they would have to fine tune it a bit more.

Dhaval:
Wonderful. What was your biggest learning lesson in terms of finding the product market fair, especially with AI capabilities? When was that light bulb like, yeah this is happening. I know you are an AI first product, but uh, just in terms of okay, yeah, this is where we are starting to see the fit. What was that? How, what was the learning lesson there?

Abhi:
I think I, a lot of it was like market being at the right place at the right time as it's the case most of the times having been in the AI space for the last five years building like so I've been working on this AI chatbot tool for individuals and influencers. But then the tech wasn't there at that point, to create any sort of. Custom implementation, you would have to train tons of data and even then the responses wouldn't be anywhere close to what the user would expect. So having been through those struggles, and then suddenly when open AI release, GPT 2 GPT 3 I could see the remarkable differences in the output quality. And that's when like I said, I realized, okay, well this could be packaged into a much better, bigger product for a lot of these copywriter and marketeers. And to be honest with you, that that was, that seemed like the first, logical use case of this technology. Now you can think about a lot of other things, but at that time, I think content creation creative kind of copy generation was probably the first thing which would come to your mind, and that's how we got started with this. I think that's when it hit us. Hmm. This could actually do a lot of good things for small businesses and startups.

Dhaval:
Yeah. So for you it was more of a, just making sure that the, the idea and the insights that you had around being able. Use a large language model for the end users and being able to implement them. That idea and follow through itself was enough for you to achieve that initial market pull. There were not a lot of hiccups in terms of like your journey itself. Once you had a great idea, you had a good execution and you just found the fit, there wasn't a lot of like oh, we did this and then we had to pivot and we had to do this and that and that. None of that stuff. Am I getting that right?

Abhi:
I think for this idea, yes, but I wouldn't say that it was like a smooth journey from day one of course we had to struggle going from zero to let's say first thousand customers, that's always the biggest challenge, right? I think that part, like first one or two months, like literally had to burst our assets to actually go and get first set of earlier adopters and users. That part was tricky and I think it had less to do with the product and more to do with the marketing side, the kind of customers personas that we are targeting. And yeah, I think we just, uh, pushed through that challenge and that's how I think those first two or three months were so crucial in getting the foundation right. Building that initial set of audience and then from there on we didn't look back so from product side and market side, we didn't pivot at any point. But yes, the distribution was always challenging, in early days.

Dhaval:
Yeah. Tell us a little bit about how you went about acquiring your first thousand customers, Abhi.

Abhi:
Yeah. That's always actually zeal for most of the startups. But, we mostly used organic channels and even today we spend actually zero on marketing, as such so you know, initially it was mostly LinkedIn, Facebook, Reddit you know, these communities and groups that we targeted made up of, like mostly copywriters. Digital marketers getting them on board as beta testers inviting them to give us feedback, giving some sort of coupons and discounts. And then we had a very strong focus on email marketing as well. Like we had set up our own funnels to convert these visitors leads into retained users and probably paying customers. And then the whole SEO was like a huge channel for us, which we had invested heavily in from early days. Again, I have a fair bit of experience on, on that side so we went all in on creating content, optimizing it for seo building quite a few I would say, authority in this space, AI writing space. So that led us to acquire a lot of organic traffic from searches. Yeah, so I think these two or three things worked really well. And of course we had partnered with a lot of these early stage deal platforms like AppSumo and all, which gave us a lot of feedback and traffic in the early days, which helped. But yeah, it's, it was a mix of quite a few things, so I wouldn't say that, oh, this is the only thing which worked for us and that

Dhaval:
lot of hustle. what was the differentiation for your product? How did you compete with other such AI copywriting tools? at that time.

Abhi:
That's a great question actually. So yeah, we had a fair bit of competition, when we launched it wasn't as noisy as it is today, but, there were still like few people who were fairly established in the space. But like I said, I think one of the reasons why we did what we did was we realized there is still a gap in terms of user experience. A lot of these tools were like very cumbersome hard to use output quality wasn't good. They were using probably different set of models or training data. And we spoke to ourselves and said, well, we can do a better job at it. We, I have a product background and I, we really take pride in building delightful products, which really, you know make the customer happy and they should have a smile when they're using the product. So it's that level of I would say finesse that we want to provide in the whole, into an experience. When we embarked on it, we wanted to keep it super simple, intuitive, easy to use, and high quality outputs that people can get at a very, very, I would say, fair price point. So that is to date our value proposition the easiest most intuitive air writing platform with the most value for money for the end user. Yeah. So I think that was our differentiator. And Today that has been working very well, I would say

Dhaval:
Abhi you mentioned fair price point and high quality output, and you also mentioned training data. One question I have for you is how did you acquire your training data and what do you recommend other AI product creators do when they're thinking of training or fine tuning the models? What is. Some of the best strategies to acquire training.

Abhi:
It depends, on the domain to be honest, that you're working in. So like in our case, if you're writing copies for like, emails, blogs social media, I think it's easy to get that sort of data, right. Again, from my experience, I Have tons of, like you can say swipe files which I could use as like high quality examples of what I would expect the AI to generate and then tweak it a little. And do lots of permutations and combinations in terms of the kind of prompts you wanted to generate for the ai so I think that experience helps some domains have that knowledge freely available. And the good thing is with the GPT, you don't need that many examples of training data. It's, it can work with few examples as well. So we use that, we leverage that knowledge and some of our experience. But I think if you're working in different domains, let's say maybe finance or healthcare I think that's where it gets a little tricky because you need like very specific, factually correct, data and which is very, very customized to a given client or an industry. I think that's where, again, if you don't have, let's say, roots in that space, it would be very hard to go out and acquire that training data or knowledge base

Dhaval:
you got acquired by Copysmith. And that's a public information. One question I have for you Abhi, is how did that acquisition happen? Like did you choose to get acquired or were you revenue positive? And the Copysmith team wanted to expand their operations. tell us a story about that acquisition.

Abhi:
Yeah, it's interesting and amazing folks we have with the with us now. But yeah, I think it was mostly to do with how the vision and the products aligned, to be honest. And, the team at Copy Smith is incredibly good. the the CEO Shegun again, one of the very sort of visionary founders who has a very clear vision of the industry, the way the market is headed and what he wants to do. So it was more like a complimentary sort of brand and product for Coppersmith to add to portfolio. they have their own focus. We have our own focus. So it's it's like collective of AI. Writing or content generation products that we are building as part of this umbrella entity. So it sounded like a very natural match, for us. And again when it comes to scaling it, we would really benefit if we had somebody like coppersmith and bigger sort of people helping us and providing us the resources that we need. So yeah, so that's how it worked out. And again, good people to have on your side. I would say. Any day.

Dhaval:
That's awesome, man. Congratulations on that. Final question. For AI creators who are just starting on this journey. They're just getting started. What is the future look like in your point of view? What is that?

Abhi:
Yeah that's a great question Dhaval and to be honest with you, I think about it a lot and it's hard to open your phone or Twitter or any other app and not see any openAI chatGPT or AI related news these days. It's everywhere. And and my view is like it's getting noisier and noisier. it's a lot of every other day there's like hundred different companies and products which are coming up so my fundamental belief is that, again, going back to the first principles, if you are somebody who wants to venture into this space, think about the fundamental needs that will never change for the end user. Whether it is the customer or if you're talking about businesses or enterprises, it's about those industries. So if you're looking at it from that perspective, I feel aI will be mostly a feature in those products, existing products and services. So it'll be an embedded kind of experience. Which most of the products will provide at some, in some shape or form. So how you can, build something which can compliment their existing products is what I would do. If I were looking at this space or if it is something which is like completely new, which doesn't exist today. And that can only be powered by this kind of technology. Then that is another lens to look at it, right? So something like let's say copywriting and content generation, which there was no other company which was doing this. So it made sense to create something like this but let's say if I'm a finance company or maybe a healthcare company, I want to have AI embedded in my existing, suite of services, then I think it makes sense to do something for those enterprises instead of establishing, let's say, standalone features or products.

Dhaval:
Thank you. Thank you so much for sharing your knowledge being here. Looking forward to staying in touch with you. Hopefully invite you back to the podcast once you build your next startup or grow this startup and share your learning lessons. Thank you, Abhi.

Abhi: Thank you. Dhaval thank you.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI game that attracted 100K users in its first week of launch.
28 mars 2023· AI Product Creators
Nick Walton is the CEO of Latitude, an AI-gaming company known for creating the first-of-its-kind AI-generated text adventure, AI Dungeon. A builder at heart, he created the first version of AI Dungeon in early 2019, a revolutionary experience that had 100,000 users in its first week after being launched. Along with continuing to develop AI Dungeon, Latitude is re-imagining what games could look like with AI and is working on a platform that will enable creators to make their own unique AI powered games. In today’s episode, We discusses the early stages of his startup, focusing on the challenges they faced and their approach to overcoming them. They also explored optimization techniques for model deployment and AI provider evaluation. Nick emphasizes the importance of focusing on fundamental human values and needs when developing AI-driven products, rather than novelty and short-lived appeal. Tune in to hear Nick's insights and experiences in building Latitude and how you can apply these lessons to your own business.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Nick Walton:

• LinkedIn: https://www.linkedin.com/in/waltonnick/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
[00:01:00] - Introduction to Nick and AI Dungeon
[00:03:45] - Initial challenges faced by AI Dungeon
[00:09:03] - Energy expenditure in the early days: dataset and fine-tuning
[00:10:05] - The role of ML operations in AI Dungeon
[00:11:44] - AI Dungeon: A game company that uses AI
[00:12:05] - Adapting technology and unit economics
[00:13:03] - Balancing user expectations and cost
[00:14:02] - Optimizing operations and advice for aspiring AI entrepreneurs
[00:16:23] - Tapping into instinctual user needs
[00:17:43] - The future of AI in gaming

Transcript:-

Dhaval:
Welcome to the podcast, Nick. Tell us a little bit about your product.

Nick:
Yeah, our main product is AI Dungeon. So it's an AI powered role play adventure game where people can jump into variety of different AI experiences and make choices that result in a fun and interesting story. So, traditional text adventures where there's a limited set of options you can really select and you go down a path that a developer pre-imagined with AI dungeon every time the story is unique written by an AI. And so you have freedom to create all kinds of different stories that the developers us would never have imagined possible.

Dhaval:
That's amazing so. Is this like a video game? I've never played Dungeon before. So what is this for people who have never played Dungeon? Is this a video game? Is this like a regular board game? Tell me a little bit about that.

Nick:
Yeah, it's like a video game like a classic text adventure game where the game shows some story of where you're at and you type actions and then there's a result. It's a little bit more towards the end of a creative sandbox than it is a really structured game with lots of mechanics. So it's more on the story roleplay side.

Dhaval:
Got it.Wonderful. So how do you serve your customers? Is this a mobile app? Is this a web app?

Nick:
Yeah, We are on mobile and on the web both.

Dhaval:
Wow Okay. And is there a specific segment of your customers, of video game players or other players that you address with your game?

Nick:
Yeah, I think for ours, we're especially targeting kind of role play gamers. So gamers who are interested in role-playing games like DnD and tabletop ones, as well as our traditional game RPGs, like Skyrim and things like that. Where you get to kind of decide who you are and what direction, and what actions you want to take in a more open environment

Dhaval:
wornderful Tell me a little bit about your own journey. Are you a game developer or are you a techie who got interested in games? Or are you a business person who decided to try a new idea? Tell us little bit about your personal background.

Nick:
Yeah, definitely the techie category. So, I come from a machine learning background. Before I ended up doing this, I spent a couple summers at self-driving car companies. I was going to go work at Aurora. And this side project that I'd been working on AI dungeon, just kind of took off on the internet and I realized there's something cool here that people are excited about. So, I pivoted from where I was going to be a founder. So I have learned lots about kind of game industry and the business side since starting Latitude. But going in, I just had the kind of tech machine learning AI background.

Dhaval:
And Is Latitude already generating revenue? Is that a revenue positive organization?

Nick:
Yeah, we just reached profitability pretty recently.

Dhaval:
Wow. Congratulations. That must be a big deal. So if you could share any information on your revenue or your number of users, or any capital that you may have raised that will help us understand the context of your product.

Nick:
Yeah. We've raised over 4 million. Since we've reached profitability, we haven't had to raise in the last little while. We've had millions of users download AI dungeon and we have a pretty active excited player base.

Dhaval:
Wow. Millions of users profitability. And when was the last time you raised?

Nick: End of 2020.

Dhaval:
Wow, more than two years ago. So you are a fairly successful case study here in terms of building an AI product. I would love to dig in a little bit here in terms of what was your journey like in terms of building the product? So you said you are a techie, you have background in machine learning. Did you find some video game players, video game creators? Did you have a passion in this space yourself? How did you go about creating a business model and a product canvas for your capabilities?

Nick:
Yeah, it's kind of funny because I think I very much came at it from like hacker creator who got kind of surprised. And so I feel like it's taken me a while to catch up on some of these other things around understanding our product, what it is kind of game industry space. So the way it really got started was just at a hackathon. I was just playing around with the smallest version of GPT-2 had just come out. So it was a hundred million parameters, a thousand times smaller than the language models of today. But even then I could see how this kind of AI's ability to do dynamic storytelling was really fun and interesting. So I got super obsessed with it. Worked on it over the next like nine months. I would have my friends play and I'd just like watch them and see kind of what their experience was, what was challenging and what made it more interesting. And so put it on the internet, not expecting that it would really launch off and be a product. And, so I was kind of surprised of scramble after that to figure out how do we make a product out of this so that people can continue playing. Because when it first launched, it was on Google Colab. And there was so much traffic. I knew that they could not let it be there for very long. So, I wanted to get an actual app and product place where people could continue to play it and not rely on Google Colab's GPU generosity for more than a little bit.

Dhaval:
Wow. Colab, for people who are not aware of is like a script writing tool. Is that right? It's like one of those web editors in which you can write code. Is that true?

Nick:
Yeah, the thing that was cool about it is Google generously lets people use it and run on their GPUs for free. And they have paid plans as well. So it was a way that. People can run machine learning models and try them out themselves, and so that's how AI Dungeon was first distributed was just as code that would run in these Google Colab notebooks.

Dhaval:
Wow. So what was the journey like for you when you finalized the product and you are ready to launch it? Did you partner up with other business co-founders or are you the sole founder of this product? Tell us a little bit about your partnership.

Nick:
Yeah, once the game really started kicking off and I realized there was something interesting here. I pulled in my older brother who's worked at a bunch of tech startups to co-found a startup around this with me. So for the first couple months, it was just us with a couple people advising or helping out here and there. And then we raised our first round of funding and got kind of initial team together. And then we started going forward from there. So that's kind of what it looks like.

Dhaval:
Got It. So when you started this, you built this on top of GPT-2, which was a thousand times smaller than what we have now. And when you say a hundred million parameters I'm sorry, 10 million, did you say 10 million?

Nick:
Yeah. It was the smallest version of GPT-2. The very first version, the version that took off was a billion parameters. So it was about a hundred times smaller.

Dhaval:
Got it.Now, did you create a lot of custom capabilities on top of the foundational model or was that built in and you just used it in a clever way?

Nick:
Yeah, we did fine tune the model on second person kind of story adventures so that we would have an understanding of what that format looks like. Especially because there's not a huge amount of that kind of text format on the internet. And so I don't think the base models had a great understanding of what that was, especially earlier on when they weren't as smart as they are today.

Dhaval:Got i t

Nick:
There were also a lot of things around how you manage what you feed into the model, what it comes out, how you alter the output so that it would feel more like a chat experience where the AI is telling you a story rather than stopping mid-sentence on the way

Dhaval:
That was also a big effort in your engineering, in your output engineering for the prompts. Is that what you're saying?

Nick: Yeah. In the early days. Getting that piece right.

Dhaval:
Yeah.Where did you spend most energy in the early days? Fine-tuning the model, getting the data, training the model with the fine-tuned expectations. What was that like? What was the energy expenditure in the early days like?

Nick:
Yeah. So Pre-launch it was a lot on the dataset. Changing the data set, like kind of getting it into the right point and fine tuning it so it worked well. After launch, a lot of people nowadays use language model APIs, but back then there was not language model APIs. We had to spin up clusters of hundreds of GPUs serving GPT-2 and people weren't really doing that yet. And so we spent a lot of time just managing those, trying to keep them up, trying to request more and more GPU quota from Amazon AWS because they don't like it when you want a ton of GPUs really quick. And a lot of the early days was just managing and adapting to the huge amount of traction we had early on.

Dhaval:
Interesting. So, would you say that falls in the category of machine learning operations, orchestrating all the different pieces together to run an app together that has a foundation of ML? Am I getting it right? Was it ML ops or something else?

Nick:
Yeah, I think you could call it ML ops.

Dhaval:
Okay. And then your strength is in building ML models or is that in making the business case out of the ML models? I believe it's a former, but I just wanted to unpack that a little bit.

Nick:
Yeah. I think we've even kind of changed how we've thought about it over time. So we used to consider ourselves more of like an AI startup. And I think in a lot of ways we are, especially compared to some, but at the same time we don't really think of ourselves as an AI startup anymore because I think that's it's easy to think of like the hammer you have relative to the nail. Really what we care about is helping users have a rich, like meaningful role play experience where they can jump into the shoes of a character, be someone they're not in normal life, and go on you know exciting fun adventures. And we use AI to enable that much more effectively. But AI is not the goal. It's just one of the big tools we use. Like If there's lots of other work we do to enable that kind of fundamental human desire that aren't AI. And so i think now like we certainly have a lot of things we've learned about how to leverage AI models to do that, but I think we're especially focused on that user desire and need and fulfilling that and using AI where appropriate.

Dhaval:
Yeah. Interesting distinction. You are not a AI company, you are a game company that uses AI. Interesting distinction.

Nick: Yeah.

Dhaval: What was the journey like for you in terms of experiencing all the growth? What changes did you make in the fundamental technology once you reached profitability or once you reached a certain level of confidence in your product?

Nick:
Yeah. I think one of the hardest things was, has always been the unit economics. So, anyone who has worked with these large language models, especially especially GPT-3 size models, knows that they're very, very expensive and especially for a consumer use case and a game where people don't expect and want to kind of get limited and say like, you can't play anymore because you don't have enough. I think adapting to that and user expectations was very, very challenging and figuring out how to make the unit economics work so we could provide a unit consumer experience at a price that was within consumer ranges because you can't do 50 or 100 dollars a month for consumer subscription because that's out of the ballpark of what the majority of consumers can afford or imagine paying

Dhaval:
And where did you acquire that? How did you come to a place that was a happy place for you in that area? A lot of testing, a lot of iteration or something else .

Nick:
Yeah. A combo of playing around with kind of the structure and our monetization and pricing structure over time, as well as just working on the how we deploy the models. So for our free models, we deploy them ourselves on GPUs because that's how we can get the cheapest cost possible so that we can support free to play, which I don't know how many like yeah it's hard to do free to play with kind of the cost that go into these models. And then with, I think the other thing is just testing and evaluating different AI providers and just finding ones that match. You know needs and what we can do for our users. So I think those are kind of the combo of those things. We've just iterated on as well as there's some other optimizations in terms of like and how and when you call the models, right. Like if you do one call with multiple generations versus one call with only one generation, like you can cash generations and there's some things there you can do to get some gains as well.

Dhaval:
So it has been a lot of optimization in the operation and the unit economics. What advice do you have for game creators or for product creators who want to use the new technology to build a product? Like, any advice for new and up and coming entrepreneurs using AI.

Nick:
Yeah. So, one of the things and I mentioned this earlier, one of the things I see and we had ourselves for quite a while was I think a lot of products and things you see are somewhat like people love trying them out and they play them for a few days, or they like play around with it a little, but then there's like, there's not staying power. It's not really a product that someone will use long term as much as it is kind of like a fun toy they play around with and set aside. And I think the generative AI tech lends itself very well to like, oh, look what you can do. And it's kind of fun. But I think the challenge in the space for people who want to build big businesses will be what is the like fundamental human value that you're solving here, Right? like One of them mental models I like to use is like, what's the like caveman equivalent of what you're doing, right? Like for us it's role playing where for thousands of years people have played in a way where they step into the shoes of another character and kids do that kinda like, it's an instinctual human thing. I think if people can understand that really well, like a good example is chatbots. Right like I think that there's a lot of AI companies doing chatbots, but I think the problem is a lot of the ways they do chatbots is as a toy and as a novelty, not kind of a fundamental human need, right? And so I think companies who will be successful with those won't approach chatbots as kind of that fun to chat with this a thousand. You know, these a thousand different interesting chatbots, but rather it would be It will try and match the fundamental human experience of like building a relationship with another being and like the progression in the relationship, how you start more guarded or unfamiliar and you develop greater kind of like emotional connection over time and feel this closeness. And I think that will happen with only one or a couple characters rather than this wide swath where you try one, set it down, try another, set it down. So anyways, that's one thing I've thought a lot about in this space.

Dhaval:
Yeah. So going back to the value proposition or any product creator who's interested in creating an AI product using some of the fundamental foundational models, it's not the fact that you are doing something cool or something novel that's going to make you successful. It's a fact that you saw a real pain point of your users. That's gonna make you successful.

Nick: Yep yep exactly,

Dhaval:
Yeah. And I love how you have have this analogy of tapping into the instinctual user needs. That's even like more powerful than just a simple pain point, right? So if you can tap into those instinctual needs as a product creator, you increase the retention, the stickiness, the relationship with your user needs.

Nick:
think the pain point model works well when you're building a tool. Right. Like I think pain points are like humans have been using tools for thousands of years, right? Like fire and weapons and all these things. And so I think it works well there I think it isn't always as applicable for consumer things that aren't a tool but are are more of an experience or fill some other need besides helping you accomplish something else you already want to do.

Dhaval:
Yeah. Very interesting distinction there. Thank you. Thank you for sharing that. Wrapping it up, I'm wondering what do you think is the future of your product? Where do you see it going?

Nick:
yeah so , two things that we are really excited about in the long-term, enabling with AI in the gaming space. One is creating and enabling people to create games that are dynamic and alive in a way that they haven't been before, right. So past games, everyone goes on the same quest and everyone is kind of the same hero you know you have three options you can choose, right. So there is kind of projection of choice of choice But your actual choice is fairly limited. And so we imagine games of the future where you have this like living dynamic nature where you might have individual quests that are just for you. You might have conversations with characters that you build deep, meaningful relationships, right. And so I think there's this whole class of games that we've imagined as humans, but we've never been able to make before. And we'd love to see those come to reality. The second thing that we're really excited about is the ability of AI to democratize game creation. So you think about what the digital phone did for creators who wanted to make video or picture content right?. They spawned entire platforms platforms like Instagram and and just this wealth creation that before could never happen because it required kind of a professional Company level. And I think we're going to see the same thing with games where games have traditionally been somewhat hard to make. And I think AI will lower the bar for that. And so we'll see a lot more. Everyday people be able to have a cool idea and be the kind of creative director with an AI team to bring it to life.

Dhaval:
Wow. I love that. I love that analogy that you just made. Thank you so much for coming out to our show Nick. And I learned a lot just from our brief conversation. Looking forward to have you back on the show once you have a few more learning lessons to share with us. Any last thoughts to share with us?

Nick:
I think there's a lot of kind of like fear and worry with some people in kind of the AI space. And I think technology, new technologies are always hard. The way I see the future is not that AI really replaces people's jobs, but rather, I think the best example is with chess Right. So like for a long time, humans were the best at chess, and then a computer finally beat the best human at chess. But now the best kind of competitors in chess are not human or AI alone. It's human AI teams that work together. That are more successful than human or AI by themselves. So that's kind of how I see the future of AI and technology is not that AI is like replacing us as humans but rather that it's augmenting us so that we as a team can do much more than either can alone, because I think human will always have huge advantages in how we can think and our flexibility and our ability with ambiguity that machines won't. And I think machines can continue to kind of automate the less ambiguous You know sides of creation, enabling us to do more interesting things.

Dhaval: Thank you. Thank you so much, Nick.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI product to help writers beat creative block and generates over $750k ARR.
27 mars 2023· AI Product Creators
Amit Gupta is the Founder at Sudowrite. He is the former Founder at Photojojo, He sold Photojojo at 2014. After selling his first company Photojojo in 2014, he started traveling, writing sci-fi, and more recently, building Sudowrite. In today’s episode, We discussed the benefits of using AI to assist writers, including increasing their productivity and providing creative inspiration. He shares insights on building AI products using pre-trained models like GPT-3, as well as the challenges and costs involved in training custom models. Amit also reveals the future vision for Sudowrite, aiming to help writers create more content and monetize their intellectual property, effectively turning them into franchises. Amit shared about building on top of pre-trained models vs training custom models, the importance of personalization. Tune in for valuable insights into the exciting world of AI-powered writing tools.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Amit Gupta:

• LinkedIn: https://www.linkedin.com/in/amitgupta/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introduction to Amit Gupta and Sudowrite
00:02:37 - Sudowrite's capabilities and how it helps writers
00:04:12 - AI writing assistance: Assisting vs. Replacing writers
00:08:41 - Sudowrite's goal of supporting writers in their entire journey
00:09:17 - Building on top of pre-trained models vs. training custom models
00:10:58 - Fine-tuning pre-trained models for new AI products
00:11:42 - The future vision of Sudowrite: Turning individual writers into franchises
00:13:08 - Sudowrite's seed round funding and bootstrapping experience
00:14:11 - Inspiring new AI product creators to start their own projects

Transcript:-
Dhaval: This founder built an AI product to help writers beat creative block. He's a former writer himself, and in this episode we talk about how to approach building AI products in 2023 and beyond. Specifically, Amit shares his journey of how he launched in August of 2022 and generated a profitable company within a short amount of time with long enough runway. To allow his team and himself to chase the product vision without having to raise multiple rounds of capital. Amit Gupta and his team are building. AI writing tool called Sudowrite. He's a former founder of Photo jojo, a product that he grew to 10 million a year and sold in 2014. After taking a break for a few years, he launched Sudowrite in August of 2020. An app to help create a writers use AI to beat writer's blog.

Talk to us, Amit. Tell us about your product, your market segment, your ideal users. I would love to hear about it.

Amit:
Sure. Our product is called Sudowrite and we're building an AI writing tool for creative writers, people who are writing novels, screenplays, any kind of creative story-based narrative content. And today our audience is those individual creators, and I think in the future, We really want to create a story engine, something that helps anyone who wants to tell a story. So that could be any of the people in the world who always thought they had a novel inside of them. It's too much work. It's too hard to get that novel out. We want to make it easier.

Dhaval:
Very cool. Tell us a little bit about how you got onto this idea. I know you are a very prolific writer yourself. How did you come up to wanting to solve this problem? And tell us a little bit about your background that you were willing to share with us.

Amit:
Sure. Yeah. I've started a couple companies in the past. The last company I started, I sold in 2014. Then I decided to take some time away from Tech. I left San Francisco, I traveled and I really decided I wasn't going to do the same thing again. I was going to try something different. So I ended up writing and I spent a few years writing fiction, which is where I met my co-founder in a writing group. And as we were writing fiction, we were bemoaning how much harder it was in running startups because, Man, you get no feedback. It's in a startup, you launch something, right away if it's working, your users will tell you, you'll see it in the numbers. And with writing, it can take weeks or even months to get a few pieces of subjective feedback on a longer form piece. So it's a very different world, much different kind of Learning curve. And we wanted to find a way to bring some of the rapid iterations and the ideation, some of the collaborative elements of working in tech or on a startup to the writing process.

Dhaval:
Very interesting. What was the process like for you to build your MVP? Did you start with wanting to build an AI first creative workflow, or was it writer first and then add AI to it? Tell us a little bit about how you did your MVP.

Amit:
Yeah, we were definitely AI first. My co-founder James started playing with GPT-3 when he came out a couple years back. A few years back. And originally we weren't planning to start a company. We were just making a tool for ourselves and for our writing group. And then we shared it with some other writers that we knew and respected. And we just got incredible feedback. So we kept tinkering with it. If I think about a year before we decide to really turn it into a company, It was for that year just a hobby project. A lot of writers that are we knew were using it. And we didn't focus on a lot of the features that most writing tools had. We focused on the things that were unique Sudowrite, the places where we could bring AI to the process and do something new.

Dhaval:
Very interesting. Did you build on top of chat GPT-3 or was that a private beta at that time? How did you get access to that? Tell us all the juicy details of how did you integrate AI in your workflow? And what was the product market fit experience like that like for you?

Amit:
Sure. So this was back in 2020. GPT-3 had just come out. It wasn't widely available. I think we DM Greg Brockman to see if we could get access and we just wanted it to play with as writers for our writers' group. So he gave everyone in the writers' group access, and started playing around with it. Eventually we started building on top of it.

Dhaval:
So where are you in terms of your product? Do Have you launched your MVP do you have revenue? Do you have certain number of users? Yeah. Tell us a little bit about that.

Amit:
Yeah, product-wise, we started building it in. I think like August or July of 2020. And we started letting people in shortly after that. We started charging people, I think last year or late the l year before, not that long ago. So it's been on the market for a bit now. And I think it's grown steadily throughout the history. As soon as we started charging for it, people started to sign up. We don't have a freemium options it's all paid. And we've really kind of seen it hit product market fit in the last couple months, or at least what we think is product market fit. It's in the past couple months where I think we've seen a lot of people signing up, a lot of people kind of entering the community, asking for things, suggesting things that are rate where like it no longer feels like we're pushing it up a hill. It feels like we're chasing the ball down the hill trying to catch up. So it's a, that's a cool feeling. It's a very quick reversal.

Dhaval:
Yeah. And do you have any numbers that you can share with us? The number of users you have, your revenue and daily active users or any of that data points? Cause we do have investors also listening to this. So, perhaps they can connect with you if you're willing to share that. Totally up to you. I understand that.

Amit:
Sure. Yeah. I don't wanna share too much, but we did share recently that we hit 500K in ARR in December, and I think we're above 750k right now.

Dhaval:
Wow. Wonderful. Do you have any data points around the number of users you have or market shares or rather market segments that you are going after? Any of that information?

Amit:
Yeah. In terms of market share, I think we obviously have hit like a very, very, very tiny fraction of the market so far. These are really the early adopters. The folks that are willing to experiment with new technology. I think there's a lot of grounds still to be covered. A lot of things, frankly, we have to improve about the product before it's really going to be something that someone who is used to a very polished writing experience is going to want to use. But today it's useful to a lot of people. And I think the market is You know all these people who consider themselves writers, but like I said, in the future, I think that almost anyone who reads has a story in them. So there's a lot of additional people we think we can bring into the market.

Dhaval:
Yeah, writing is such a fundamental thing, right? Once you figure out writing, you can create many other things on top of it. So I have like two questions that come to my mind. I'll ask the first question first, which are your writers using your product for instance, notion, like it's designed for short form documents, type of work, Google Docs notion word or are they using your product for writing? Like find five hundred page, a thousand. Novels or short documents? What type of short stories or long stories do you have a specific workflow in mind for writers? And I'll let you answer that. I have another question about product management. I'll get into that in a second question.

Amit:
Yeah. So I think the place where we differentiate is that we're really heavily focused on long form. And that means that our writers are people who are writing novels or screenplay stuff that can go up to like a hundred thousand words or even longer. That's something that's very challenging to do with today's AI models. It's something that requires a lot of kind of like work behind the scenes to condense things, summarize things to, to make it work. But I think where a lot of the value lies because it's very easy to generate short snippets of copy or taglines. It's much harder to create something Longer and more meaningful and substantive. But that's where we're working towards.

Dhaval:
Interesting. And this lends my second question, which is in the product experience, in the customer journey, in the user journey, how did you come to know , this is for AI product creators who are trying to infuse AI in their product. So this is for them, for your product. Amit when you were building it, when you were trying to find that product market fit. At what point was the moment when you decided, okay, at this point in the user journey, we'll be infusing AI . Tell us a little bit about how did you decide at what point in customer journey was the AI part of.

Amit:
Well, Sudowrite is a AI first product, so there's really no reason for it to exist without AI today. And I think so even in the onboarding, you'll start to use AI. You'll start to experience what it's like to write with AI. It's from the very first moment after you sign up.

Dhaval:
Got it. So as if a writer wants to write a book and right away you'll be like prompting them to get information data points that allows the product to support them in their journey.

Amit:
We'll show them right away, right after they sign up, what AI can do for them in their process. And I think the eventual goal is that we can help with the whole process from ideation to first draft, to revision, to publication, and just be able to be there with every step in the process. And I think some writers are really good at various pieces and they hate other pieces. So we want something that can help take the drudgery out of writings, wherever that drudgery comes for each writer. because it's different for everyone

Dhaval:
That's interesting. Yeah. I've talked to many of the AI founders in the last few months, and this is the point where I hear different opinions every time, which is, if you're a new AI product, if you're interested in creating a AI product, right? You're using one of the pre-existing tools, right? Where do you find that personalization? Do you build your own personal models on top of? Pre-trained models like chat GPT-3 or do you go a hundred percent custom? How do you find that differentiation, etc. What are your thoughts on building on top of pre-trained models versus training your own models and building a groundbreaking product from ground up? Where do you like and what are your thoughts there?

Amit:
Yeah, it really depends on the company what stage they're at, what second on the market and what kind of AI is useful for them. I think with, for instance, stability, there's a lot you can do as a relatively small company trading that model to do interesting things for you because of the way they've sharded it down in size, and I think with large language models, that's a lot more challenging to run a large language model like GPT-3 requires a lot of infrastructure investments just to have the machines capable of running the thing. And so it's quite expensive to just run inference before you even get to the concept of like training it, which could cost millions. So I think there's definitely a lot to be gained from training custom models, but at this point in the evolution of them they're Iterating and developing some rapidly that a lot more of the gains come from things like fine tuning or developing models that sit on the layer above or below GPT-3 or otherwise finding ways to really get the best quality out of GPT-3.

Dhaval:
So fine tuning on top of pre-trained model is the way to go for a new AI product creator. However, If you're working for one of those organizations that has the data available and has the resources available, building your own model may be an option.

Amit:
Potentially, you'd have to see how much additional value you really think that custom model will bring, and if it's worth the expense. I think even if you are able to build it and have the data, it may be the case that still doesn't make financial sense to build an LLM yourself today. Because as soon as GPT-4 comes out, it's going to leapfrog whatever you've run and then you're going to have to train a whole new model to just to keep up with the new kind of state of the art.

Dhaval:
Very interesting. Okay, tell us a little bit about your future vision of the product. Where do you see this going? Where do you see Sudowrite going ?

Amit:
So today I think it's very useful for Sudowrite very useful for people who know how to write and who want to write faster, who want to get unblocked, who want to keep in the flow. And those writers come to us and they tell us that PseudoWrites able to help them double their word count so they can publish twice as much stuff. And they're very excited about that. We're in a world where as soon as you finish watching a TV show on Netflix, you want the next season. And the same is true of books. So as soon as a reader finishes reading a book, as soon as a writer has published a book, the next day they're hearing from readers that they want the next one. And of course, you can't deliver that, but I think in the future we want to create a system that helps the writer deliver that helps them. Serve their readers, produce more work, produce more artifacts aside from the novel that take place in the world and in the story that they've created, and basically find new and interesting ways to monetize their IP. In the same way that bigger pocketed creatives today, like the Marvel Universe, or Star Wars or whatever, are able to like create a lot of storylines in the same world, a lot of different product offerings. We want to bring some of that power to the individual.

Dhaval:
Wow. Turning individual writers into franchise. Wow. That's a powerful vision. Amit, have you raised capital? Is this a bootstrap startup? Have you raised a seed round and if not, are you planning on doing it soon?

Amit:
Yeah, we raised a seed round late 2021 $3 million. We raised it for primarily from individuals, so people who are in the creative fields, in other related fields, and we have no immediate plans to raise more money. We have quite a bit of runway.

Dhaval:
Would you say that a AI product creator who is interested in getting their feet wet and want to build something bootstrapping to validate the idea and validate the hypothesis may be the best way to go forward? Or is there too much a bar? High bar for them to do that. Trying to bootstrap Were you at all at any point bootstrapping this and tell us a little bit about that experience?

Amit:
Yeah, it was Sudowrite bootstrap for the first year. I think there's absolutely no limitation on anyone who wants to get started with generative AI today. Building a startup from the ground up with just one person and your spare time, I think the tools are incredibly powerful and there's a wide open field for so many different solutions that can still be made.

Dhaval:
Thank you for sharing that. That's very inspiring to hear because you're a hundred percent right. There are so many remain to be optimized. And it's up to the creative potential of the individual who's trying to create that product to decide where they want to go after what they wanna go after. Yeah, that concludes all my questions. Thank you for being so on the point, man. Thank you so much.

Amit:
Yeah, sure. Glad to do it. Happy to meet you.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
How they built an AI Product where you can discover and generate art to 100x your creativity.
23 mars 2023· AI Product Creators
Coco Mao is the CEO & Co-Founder at OpenArt. And John is the CTO & Co-Founder of OpenArt. OpenArt is an AI-native social platform, inventing a new paradigm and tools for expression and entertainment with generative AI. In today’s episode, They share how they're helping artists and designers 100x their creativity using AI-generated prompts, and their journey from ideation to launching a successful product. The also share their journey in the world of AI and how they built their successful AI-powered platform. Discover the power of leveraging pre-trained models, adapting to new technologies, and focusing on user experience to make a difference in the AI industry. Don't miss their valuable advice for non-technical founders and aspiring AI product creators.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Coco:

• LinkedIn: https://www.linkedin.com/in/kechunmao/

Where to find John:

• LinkedIn: https://www.linkedin.com/in/johnqiao0618/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introductions and Open Art's mission
00:01:54 - Empowering creators with AI-generated image prompts
00:02:41 - The target audience and the line between empowering vs. replacing creators
00:04:13 - The importance of human creativity and AI's role in supporting it
00:05:13 - The customer journey and how AI gets infused into the user experience
00:07:48 - Business results, user growth, and revenue
00:08:53 - The journey from idea to launch and funding
00:10:28 - The journey from pre-seed to product launch
00:11:07 - The goals of their AI-powered platform
00:11:47 - Evolving audience and focus on creative workflows
00:12:28 - Product market fit lessons learned
00:13:59 - Quick product creation techniques for AI product creators
00:15:35 - Advice for non-technical founders in the AI space
00:16:35 - Upcoming API release and market focus
00:18:07 - The importance of community in AI product creation
00:19:19 - The differentiating factors for AI product creators
00:20:37 - Future vision for OpenAI and personal growth
00:22:17 - CEOs that inspire Coco and John

Transcript:-
Dhaval:
With me, we have Coco and John. They are the founders of Open Art, and we'll start off with the question on what does your product do, what problem does it solve and who does it serve?

Coco:
Thank you for having us here. So in one sentence, open Art is a cutting edge platform where you can discover and generate AI art. Our website, openart.ai has been rapidly growing since our launch in August, 2022. And our long-term goal is to build an AI-powered workflow for creativity. If you think about creative tools today like Adobe Photoshop, Figma, or even Canada, they help you design or create something once you already have a rough idea, but they don't really help you come up with new ideas. You'd still go to sites like Pinterest or be hands check out other people's work to get inspired. However, generative AI can really come up with new and exciting ideas, for you and or with you like no other previous technology could before. So on our platform, we empower users to unlock their creativity with AI by one, generate images based on prompts. And second, train their own personalized AI models. Just as an example, so a fashion designer uploaded her past fashion sketches and trained her own fashion design AI model. And now she can actually immediately get hundreds of new designs waiting minutes. because these designs are based on her pastwork, but entirely new, created by ai and this magical process, really 100x her creativity. So we're really excited to let more people 100 x their creativity with the workflow we're building.

Dhaval:
Very interesting. So you said you launched in August, and this is a workflow product to empower creators with, creative graphic prompts. Did I get that right? Image prompt. Is that artistic prompts? Who do you serve? Like, is it, you mentioned fashion designers. Is there a specific segment within. Creative industry that you have nailed your product for.

Coco:
Yes, great question. To be completely honest, we're still in the process of nailing down very specific target audience. But I would say right now, we are, our audience are creative workers like artists, designers, including, like the fashion designer talk about, but creative workers who have like ideation phase.

Dhaval:
Very cool. Yeah. There is a common theme among the AI product founders, which is if you empower the creators, then your product will go a long way instead of if you trying to replace them. Where do you draw that line? Where do you draw the line of empowering versus actually taking over a little bit of their creativity? Where does that line stand for you?

Coco:
I think it depends on how people use your product because I think the way I think about it is that for for creative workers like artists or designers, like.At least our product, they can really use it as a workflow tool that really 100x their creativity. And we're constantly talking to this artist and designers how our workflow could actually help them monetize more like help them get more clients. So the way we work with them is very, very collaborative. Whereas I do think in terms of replacing artists I can see like in some use cases, for example, if you don't have artistic skills, let's say you are a writer and like you need some illustration, so perhaps before you don't have good tools to get any illustration. But now you can use our tool to make some basic illustration. However, I do think if you need high quality illustration, you still need to go to artists. So I think the AI could replace some basic work of artists, but then like they want really replace artists.

Dhaval:
Very interesting

John:
also think about the human beings, right? So the, the most important or valuable thing is our creativity. The AI just unlock the whole potential of our creativity. So instead of wisdom that maybe 5 years, 10 years, to master like your skills to draw a picture. To paint on the wall. So the AI can help you do that. The most valuable thing I think is, of course your skills is much more valuable. But the creativity is the core part of the whole thing.

Dhaval:
Yeah, I completely concur with you on ai supporting the creativity and creativity being the most important thing for human beings, right? One thing you said is the customer experience, where in the customer journey. Is AI prompting tell us a little bit about your customer journey and where does AI get infused, like specifically in, in the user experience? Yeah, if you can share that a little bit. And how did you go about making that decision?

John:
I can give you an example from one with our user on the platform. That, that guy he has a, like a blue character called Coco, , he has like a maybe 20 to 30 image of the, that character he designed and he upload to our website and use our photo booth feature to generate a model.Then one of the special feature we had on our website is like a presets we have like a presets team in our community. With Very good at writing prompt, writing prompt for the Generative AI is really hard, right? But they're super great at that. So we have a lot of presets and that guy will just by bunch of presets and apply those presets on his characters.And you can see the result are super creative. the blue character is on different off page is in different contexts in the background. And I think he really like that because he's trying to upscale and enhance the image for lots of the results. That is really impressive thing.

Coco:
Yeah. And I want to add one thing is that after, for example, they see so many amazing results. They usually pick the ones they like, and then they can use it as a reference sheet to rejoy it. Or they can actually, if the quality is, they're happy with the quality, they just need to do some touch up on the, on the on the image. And then they can actually, that can be their final like image.

Dhaval: Wow.

John:
I think the guy also came back later and he also purchased lots of model and know, keep changing that factor.

Dhaval:
Wow. This is very interesting. The three keyword I picked up here are preset prompts and outputs, right. Is the sequence presets, and then from the presets you create prompts and then the prompt from the prompts you create final outputs. Multiple outputs. Did I get that right? The sequence.

Coco:
almost, I think so the step are like one you need to train your model and training process is super easy. You just upload some photos of the thing that you want the AI to learn. It could be, for example, an image of yourself or image of a character or image of even a consistent style. So you upload it and after 30 minutes. The model is trained. And and then you can purchase those presets that John mentioned, which are essentially highly curated prompts. And then those prompts will be applied on your model. And then the third step is that after those prompts are applied on your model, you can see like hundreds of amazing results immediately.

Dhaval:
Very cool. Wow. I fully visualize it now. Thank you so much for clarifying. Beautiful product. One quick follow up question is, you said you launched in August, 2022. Tell us a little bit about your business results. How many users you have, how many users are using your product, what is your revenue, if you're willing to share that et cetera, et cetera.

Coco:
Yeah I think last month we have around 800,000 visits on our platform. And yeah, and our SEO is really picking up, if you search an AI art generator, I think we're the top three results. And in terms of revenue we wouldn't be able to share the specific numbers, but then it is in the range of like tens of thousands a month.

Dhaval:
Wow. That's a really good MRR for someone who's, like just launched in August. Right. So six months in. Oh, thank you. Yeah. Tell us a little bit about whether you bootstrap this, whether you funded yourself, whether you raised some seed round. Where are you in that journey?

John:
Yeah I can tell a little bit story about this. It's kind of like a also like a related topic of how to find a good program marketplace. I think that overall, the timing is really important because the before open art, Coco and I actually quit from Google quit Google together, last year. And we had idea to build a marketplace for social and medium influencers and e-commerce brand. But that didn't play really well eventually. We pivot and we are very proactively looking for different ideas. I think it's Back in, August open AI DALLE-2 started getting lots of attention, right? Coco and I have a daily idea of brainstorming session, and we do it every day. We analyze different ideas, directions and see if there is something we really want to work on. So I remember it's a Friday afternoon and we were discussing about the DALL-E 2 because both of us already have the early access and we have been using it for a couple of days. And then we found a pain point that people are struggling to come up with a good prompt to generate like a high quality image. So we think we had a solution and I'll tell you a solution a little bit later. And then we immediately decided to jump on this and we build our prototype, like first of version, In a week and on the day, I think it's August the 26th. So Coco made a post on hacker news. And the title is Open Art, Pinterest for DALL-E 2 Images and prompter. Boom. There were like over 10,000 people visiting our website within 24 hours. That's how, we started at the beginning.

Dhaval:
Wow. did you say you started in week?

John:
And then tell us about what happened after that. Like what has been your journey? Like, have you been bootstrapping since? Has the customer funding you, are you raising capital? Are you in middle of it? Yeah. Where is that journey?

Coco: actually,uh, we actually did a pre seed found race last year. Earlier. Last year March or April. But it was for another idea and just like many startups we pivoted but we are thinking of raising our seed round this year.

Dhaval:
Very cool. Very cool. So you said you launched a product on Hacker News. Within 24 hours you got over 10,000 users and it wasPinterest for DALL-E 2 image and prompts

John:
Oh, the, the goal is mainly for help people to discover different AI generator images and also know every image has the prompt so they can use the prompt or get some inspirations from a prompt to generate their own images.

Dhaval:
Wow. Okay. And is this, is your audience mainly creative individuals or is enterprise tools like Figma or Canva who want to be able to support their users with creative workflows. Where, where do you focus on right now and

Coco:
Yeah, I think. I think we're constantly involving. At first when we launched, when John said, I think all we wanted to solve was a small painpoint that people have, they struggle to write prompts. So we want to have a search engine and a, like a Pinterest form to help people write better prompts. That's how we started. We were targeting AI enthusiast, but as we were building, we realized how powerful it is for creative workers like Artist and Designer to actually use it to help them with their existing workflow, to help them like come up with new idea in the ideation phase. So that's what I think we're focusing, a workflow for creativity.

Dhaval:
Thank you. John, you mentioned about product market lessons. Do you have any other lessons to share with us since you launched? You have a lot of users, 800,000 people visiting your site every month, tens of thousands of revenue. Tell us about the product market fit lessons you have learned since your launch in August. Any big lessons that you wanna share with us, especially focus on ai. How do you find that balance for other AI product creators who wanna build an AI product and be as successful as you?

John:
Yeah I think Coco can add more, but I, I can write one more thing before I mention timing is very important, right? So we were really lucky to to write on the , tailwind. So I think another one lesson learned is in this industry specifically for Generative ai, so. The change is so fast. Every few weeks there, there's something new and come to the industry and you have to act really quickly For example, at that time, August last year. The most appealing thing is the text two image, right? So people are creating, uh, different images. But the, the November, there's a new technology, like a called Dreambooth you can train like a personalized model within an hour. We quickly adapt that technology and build our own product. So I think that's the biggest lesson, I kind of learned so far. It's like that you really have to act quickly and, keep listening. The community, the industry.

Dhaval:
You said there was dreambooth. That's the product you use to customize your model. Based on that you quickly created an MVP. Any other quick product creation techniques or tips that you have for other aspiring AI product creators in this space?

Coco: I think I do have, one piece of advice for founders, especially actually non-technical founders because AI can sound very intimidating for non-technical founders. However, the thing is that a lot of, like a foundational models like stable diffusion or DALL-E 2 or GPT3 on these foundational models, they're becoming commodities. So there are so many opportunities in the application layer. Like for a non-technical founder, it's a really good thing because you don't really need to be the expert who build all of this. Like technologies from ground up. Instead, you can use the APIs, you can use like these open source projects. And the most important thing is to actually have some unique insights in specific industry. I would say that the best AI products are probably the one, that is a combination of your unique insights in the specific industry and ai. So as a technical let's say if you are a non-technical founder but you know, like specific industry so well, and you can just think about how these foundational models can be applied to solve some problems in your industry.

Dhaval:
Yeah, you had so many good gems in that last 30 seconds. I'm just gonna unpack that a little bit. So, for non-technical founders, what they can use is commoditize foundational models that has been abstracted out. So they don't have to build ML models. They can use the existing models, but what they need is they need specific experience in a given industry to be able to pull insights from that so that they can customize these models. Using other tools like Dreambooth to make it specific for that insight that they have acquired from having worked in that industry. So use the commoditized model, use your ton of experience in a given industry to build on top of that, and now you have a product that can serve other people who are in that industry. And you don't even have to be coding that you can just put it together using bunch of APIs and, existing models.

Coco:
Yeah, exactly. APIs for example, I think we are going to provide , our APIs very soon because a lot of developers are actually asking for it. So I think we will also make our service very soon as APIs so other people can even build on top of us.

Dhaval:
Very interesting. I was curious, when you, when you first release your APIs, are you going to focus on a specific segment of market or are you still discovering that?

Coco:
I think the most, the, asked for API is the. Like related to Dreambooth, it's our photo boost feature which allows you to train your own personalized models. So the API will probably be the one that allows you to train your personalized models, AI models easily.

Dhaval: Got it.

John:
Yeah. We actually have like a group of people like, who really wants the API from us. And basically, like Coco mentioned, they want API to fine tune the model because we have the best experience. And also they're asking for the crystal API to call to create images and search API because we have like over 10 million images on our website, we previously collected, and we also have the community to keep generating the new images. So those developers want to, gather a API to query the whole data set so they can use the images to somewhere else. The image itself is definitely for available for any, u usage like a commercial or non-commercial.

Dhaval:
John, you use a very important insight here, which is community. I wanna unpack that a little bit. You can build AI products using foundational models that are already pre-trained products like GPT3 et cetera, et cetera. On top of that, you can use products like Dreambooth to fine tune those models to make it specific to your industry. Now, with this two products, your technology stack is taken care of. What remains is your specific industry experience that you can bring to product creation and your customer experience, your user experience, and your community. Would you say, now that there is so much commoditization of ai, would you say that. The three differentiating factors for AI product creators is gonna be user experience, the community, and the industry insights that they have gained from having worked in an industry for long. Would you say those are the three differentiating factors for AI product creators, or is there still a need for in-depth technical, fine tuning to be able to create meaningful, valuable, net positive product?

John:
I think you summarized it really well. I think that three elements are so far as we can see right now is like a very core part for a AI company to success, at least at the beginning. And in terms of that more deeper technologies. I personally think it, it's true you eventually you have to create the technical bar. To really to make something from good, great, and excellent. Right? Excellent. It's like a copy AI or maybe a Japer ai. What they did before, they just leverage the existing tool or maybe API to build the first momentum. And regardless what resources they have at that time, but eventually they will have their core technologies to build a mode to build a the true differentiator for their company.

Dhaval:
Thank you. Thank you, John. Yeah, that was key insight for this podcast. Thank you so much for sharing that. Yeah, I think we are gonna change the gears a little bit and wrap this up. Now, do you wanna share. The future vision of your product, of yourself, and maybe tell us a little bit about where do you see you going? Where do you see yourself? Like where, what is the future for you? And then what is the future for your product? Tell us a little bit about where you see the future vision of yourself and your product. Yeah.

Coco:
Yeah, so I think the future for our product open art, I think the possibility is really unlimited. Even though I think I can give like a five year vision, but I think just based on how quickly the technology advances or like how, how much insights we're gaining every day. It could be different from where we envision today, but if you ask me today, I would say we wanna build, a workflow that can really 100 x everyone's creativity but I think in terms of details, it really depends on I think the technical breakthrough will really affect our roadmap because a few months ago we didn't know about Dreambooth and like personalizing or AI models was like, we didn't think about that. But then the technology came out and we really see there's like great value in using this for creative workers. So I think in the next few months and years, we will have like amazing technical breakthrough that will enable a lot of things we cannot imagine today. But I do think. Our goal of like 100x creativity will still be the same. And in terms of ourselves, I think we are really enjoying this journey. We really hope to raise our seed round this year, and become profitable. I think we're on a very good trajectory on being profitable. And yeah, nailing down our product market fit.

Dhaval:
Thank you. Thank you for sharing that. It was a pleasure to have you on this show. Thank you so much for being here. Any last thoughts before we wrap it? Let me ask you this question. Which CEO inspires you the most?

Coco:
I think for me, this will be a controversial one because I would say Elon Musk he is controversial, you know, but he, the thing I really appreciate about him is that he's so crazy and extreme. Like I think he's the extreme version of me that I wanna be, that I don't think I'll ever be.

Dhaval: How about you, John

John:
for me it's Steve Jobs. I think he's a really good person and he made like a huge impact on the entire society, right? So that, the one thing I really want to do because I started programming since I was really early nine years old. I build tons of shit from then. so I remember one person ask me a question, okay, you are keep building stuff and when do you stop? So this is how I tell him, okay, the time when I stop is when I work on the street. I see people are using my product and enjoy it and it, the product change their life and make their life better and make them happier. So I think that's the point when I can try to stop. Okay. So I, I made my impact the world.

Dhaval:
Thank you. Thank you so much for sharing that.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI product to help you become a super-learner.
22 mars 2023· AI Product Creators
Calin Drimbau is the Co-Founder & CEO at Broadn. Broadn is personalized learning through Generative AI. In today’s episode, Calin shares how he came up with the idea and how the product works, including the three layers of abstraction and the pipeline of information that flow through the product. Calin also shares insights on his financial journey and the challenges of integrating audiobooks into their data processing. Aspiring AI product creators will learn valuable lessons on how to approach AI product development and accelerate growth towards their vision of personalized learning.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find Calin Drimbau:

• LinkedIn: https://www.linkedin.com/in/calindrimbau/

Where to find Dhaval:

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introduction
00:01:02 - Overview of Broadn's product journey
00:02:26 - Finding the kernel of the idea and launching the product
00:05:05 - User journey and monetization plans
00:09:44 - Conceptual product architecture and data flow
00:13:30 - Building semantic search engine that feeds users valuable and summarized content
00:14:00 - Calin's background in product and his co-founder's deep expertise in machine learning
00:15:34 - Calin's financial journey and plans to raise capital for pre-seed round
00:17:58 - Challenges of integrating audiobooks into AI data processing due to copyright issues
00:19:52 - Quick experimentation and iteration using large language model APIs
00:19:52 - Importance of understanding customer problems before building AI products

Transcript:-

Dhaval:
This founder built a whole new category of ai, product known as generative learning to help you tailor your learning based on the individual context, learning goals, and learning style. Calin Drimbau the founder of Broadn. Shares how he built his AI product using. Three layers of abstraction and have, you can follow the same type of product architecture to solve your specific workflow using AI resources that he shares.

Welcome, Glenn. Tell us about yourself and your product.

Calin Drimbau:
Hi, pleasure to be here. I'm Calin. I'm the founder of Broadn. We're building a new solution for learning. It's personalized learning at its best. in many ways we're defining a new category and that is generative learning. We're using generative models to be able to tailor learning based on the individual context, learning goals, and learning style. Of users. we're very excited to be building in this space and it's a pleasure to be here and have this conversation with you.

Dhaval:
Wow, that's like personalized learning at scale. Tell us a little bit about where you are in your product journey. Has it launched? Is it, being billed? Is it in beta? Is it still being developed?

Calin Drimbau:
Sure. So we've been on this journey for about a year now, and we've launched a couple of products., the first product that we launched was a product that was doing classification on, podcast content. So we would be listening and, transcribing the text from audio and then identifying topics that are being discussed in conversation and using AI and then clipping creating automatic clippings and placing all of these into a platform for learning which was in a form of a mobile app. So that was our first product, and moving away from that, in conversations with , our users, we've learned that what they wanted to do more above and beyond getting the best clips from podcasts. They wanted to navigate and explore this content by searching. so a big problem for people in the audio space is, identifying the most valuable parts of a conversation. , and they wanted to do that by search. so our newest product that we've recently launched is, A semantic search engine on top of podcast content. and happy to speak more about that. We've launched that last week on Product hunt.

Dhaval:
Wow. Yeah. Tell us a little more about how did you find , the kernel of the idea and, how is it doing now that you've launched it? How is it received by the audience.

Calin Drimbau:
Sure. So I'm a big podcast listener, and I'm a big consumer of knowledge, if you want from books, articles, YouTube. I consume a lot of information and my personal problem was that especially with audio and video content, it's not easy to navigate, this content. It's usual. It's usually presented in a linear format. So oftentimes , when I'm looking. To consume content is because I'm trying to solve a problem or it's been, it's because I'm trying to learn more about a mental model. So having had experience building machine learning products on text, I thought, why doesn't anyone do? Processing and parsing of podcast transcripts to identify and classify what's being discussed. So that was the genesis of the idea. Beyond that, I suppose once we've launched it , in the market as I've said previously, users wanted more flexibility on how they consume this knowledge the app itself was very useful in terms of like saving a lot of time and instead of having to listen to one hour or two hour podcast, they'd be able to listen to 10 clips on the specific topics that they wanted to learn more about. A lot of the content on the platform is learning content, like entrepreneurship or product or any type of lessons and clips from , podcast surfacing and talking about these topics. But then one of the constant requests that we're getting from users is I wanna have more power in my hands to be able to navigate this content. So if we take Lenny Rachitsky's content, for example which I'm a big listener, of content too, they asked for an ability to search for specific topics or episodes or experts so that instead of listening to, the two hour episode, they'd be able to zoom in and double click on exactly the specific lesson that, that the author or the guest is, highlighting in, in that episode. So that's why we built the search interface on top of podcast content. And we've actually built it. Just on top of Lenny's content as a first drop, obviously the same technology can be deployed and adapted to surface semantic search on any podcast content. And this is one of the things that we are, we're exploring but, podcast content and searching, or semantic search, if you want, is just in our view, the first or the stepping stone in term in terms of realizing our, our vision for personalized learning.

Dhaval:
Wow. Yeah. I would love to get into the nitty-gritty of how you build the product. We'll get into that in a second. But first I wanted to dive into the user journey. So you have two-sided market. One is the listeners and the other are the creators. Are there any other sides to your market? How do you monetize this product? Is it subscription based product? Is it advertising? Yeah. Tell us a little bit about any other sides of the market you may have, and more importantly, I'm curious about is this like a one central place where the end consumers go to search for podcast clips? Or is it a service you provide to podcast creators who want to put this interface on top of their podcast so that people can consume it more effectively?

Calin Drimbau:
Yeah, good question. Obviously these are things and ideas that we've been exploring for a while now, whether we pivot into one side, a two-sided marketplace, or a one-sided marketplace for , for creators. was the questions that we've, a question that we've asked ourselves At the moment. Our vision is to create a consumer platform, not a typical marketplace if you want, because we're taking all the public content that is exists out there. So everything that a creator has deployed and it's not just audio. We started with audio and we started with podcast. But we're able to process video, YouTube and we're able to process essays and any type of blog published by, by authors and creators. in terms of the, monetization, cuz you've touched on that, the big goal is for all of these mini products that we're building and validating at the moment to form. A bigger consumer product that will have a multiple set of use cases. So indeed one of the first use cases that the product will feature is the search, or semantic search, throughout a set of podcasts. But above and beyond that, the additional use cases that we're currently working on building in, into the platform are the ability for a user to define and set their learning goals. , and, the platform, it would be able to know and understand the context of the user, and it'll generate, if you want , a dynamic personalized course. For the user. Now, the way it does that is, is a little bit dissimilar to, a typical interaction with something like chatGPT where. You'd be able to prompt the interface and ask for content. And a large language model will process that request and give back some answers to you. We're doing it slightly differently in the sense that, we have micro verse of content that we are pre-selecting, and on top of which we deploy semantic search. So this is why the semantic search, is very important because within the universe of content that is created by. We're first running semantic search, and then we're deploying, summarization and other typical, generative AI techniques to surface that content for users and, and match, their learning goals, their learning style, and the context for which they're asking that. So that will be the main product that we're building. That product is currently being built. But it will take the form of consumer subscription model to answer your question on monetization as well. Yeah.

Dhaval:
You touched on overall conceptual flow of your product. Let's, let's, dive a little bit in there. So you mentioned microverse. And then you mentioned the ability for semantic search to be sitting on top of that. So if you are to help us understand the conceptual product architecture for the people who are product creators, the audience of this show is product creators, are product creators who either want to build an AI product. And don't have the deep AI expertise or people who want to infuse AI in their existing product. So we try to unpack a little bit about how the product itself is created here not too in depth, but just a high level overview, conceptual overview. So you mentioned that, for a user's. so for your product for a user's learning goal. You would establish a microverse of content and on top of that. You would have semantic search and then on top of that you would have generative ai. Did I get that right? Please correct me on the the stack and the overall flow of information so that you can help users the end users that you have achieve their goal what is that product stack? What is that architectural data flow looking like?

Calin Drimbau:
Yeah, that's actually pretty accurate. So there's three layers if you want, and a pipeline of information flowing from this content, monitoring and extraction engine. So this is what I previously called our. Universe of content our microverse of content because we're not mining the whole internet. We're starting, for example, with entrepreneurial learning, and in this case the definition of that micro verse of content would be a pre-selected list of YouTube channels, podcasts, and articles, blogs or creator sources that will form all the data flow that, enters the the universe of content if you want. So there's a the first piece is basically a content ingestion piece. It's not a sophisticated piece in here. We're not talking about ai. It's purely subscribing and listening to a set of content. And whenever new content gets updated on these channels, we parse that information again in the second stage of the process. So now the second stage of the process, or if we're looking at it as an infrastructure layer, the second level or the second layer. Is the layer where we're deploying search. So on this layer our approach was to use the form of semantic search. We've actually built, our, and adapted our own algorithms in here. But for anyone that is new to for example, building or defining their own vector embeddings algorithms, there's already a set of pre-existing algorithms that you can use. You can either go down the route of using large language models. Open ai, the classic example, provides an API and you can use their embedded endpoint to do semantics search. Or you can go to Other providers like Algolia that have been provided providing these type of services for a long time or, pick up some of the smaller transformer models that exist on hugging face and adapt them in a way that makes sense for your particular use case as I've said, we've went down the route of picking some existing smaller transformer models and adapting them and combining them in a way that makes sense for our own use case. So that's the second part of the process, or the sec second layer of the architecture, depending how we look at it. The last layer or the final processing component is what we do with the relevant content that, we find through semantic search. And in here again, you've got multiple options. So in order to get key ideas summarized, you would, again, need to use some form of intelligent processing, right? So the simplest example again is using an endpoint from Open AI using their API and basically feeding as context, the results of the search and getting that, endpoint to summarize your results. Of course, open AI is not the only solution. They're multiple, large language models like Anthropic out there doing similar things. Or you can go down the route of a small language model as well. So all of this really depends on. Your level of familiarity with ai whether, if you're a non-technical co-founder, there's still a lot, or you don't have a technical co-founder, there's still a lot that you can do an experiment with without having, deep AI expertise because a lot of these interactions with AI models these days are in the form of a simple API query requests, and ultimately whether you need the complex AI function to be built in house or whether you just use existing models will depend on. The type of market that you're addressing. Whether it's business or consumer, what type of industry you are in, and whether you're building a small SaaS project or a venture backed business. Everything's possible this day and these days. And then choosing the right solution, if you want, really depends on on your goals.

Dhaval:
Wow. You really summarized it Well, thank you so much. This is gonna become a masterclass on how to build a semantic search engine. And feed your users small pieces of highly valuable and summarized pieces of content. I would love to, I would love to unpack, one more piece of information, which is do you have a very technical background or does your team, do you have a deep technical team, or tell us a little bit about the context and the expertise of your team.

Calin Drimbau:
Yes. So, I have a technical background. I have a computer science degree and , I've built products in, in AI and using natural language processing techniques, for six to seven years. So I personally have some experience. Building this, but I'm not technical in the sense of coding. So I don't code anymore. My expertise or deep expertise is in product. So doing anything from product discovery to execution, that's where most of my career has been focused on. My co-founder, however who is complimenting me on my skillset, has a deep. Machine learning, expertise. He has master's degree in machine learning and, also has spent 16, years or more building machine learning solutions.

Dhaval: How did you convince him to join you?

Calin Drimbau:
it was, one of those fortunate situations where we worked together, actually in in legal tech, a legal tech company building together, another AI powered solution. So we worked together for a year and a half. , and I think we've realized that we work together really well, and this is a big opportunity to use the new wave of technological advancements to redefine an industry, to redefine how we learn. We're both excited about lifelong learning and we both have the right level of expertise to build this product to live. So it was almost a no-brainer for us to work together

Dhaval:
wonderful. And tell us a little bit about your financial journey. Did you bootstrap this company? Have you raised capital? Have you raised a round of seed? Are you looking to raise it in the future? Yeah.

Calin Drimbau:
Yeah, so we've been bootstrapped so far. we're lucky because between him and me we are pretty much covering all aspects of the early stages. So I do product and design and he does development. I also do the marketing side of things with a little bit of external help. So between us, everything that we've done so far Wasn't that expensive to build if you want. It was just more our time. and I think because, we wanted to move fast in many ways, it was better for us to get our hands dirty , and start building. That's. Probably gonna change soon because given the, feedback we've gotten on the recent release and the previous release and the fact that we've built a big wait list for our products, we've received a lot of, interest from investors. So it's probably the right moment, and opportunity in time for us to raise the capital to pre seed round to accelerate the , the growth towards, the vision of, personalized learning.

Dhaval:
So One of the primary data sources for me to consume audio information is Audible, which is the Amazon Audio Bookstore. Is that, how do you get access to audiobooks in your data, processing? Is that part of your roadmap? I'm just curious.

Calin Drimbau:
It's a good question. It is something that we're considering, for the long term. It's not something that is, immediate in our roadmap if you want because with books, there's a little bit of complexity with regards to copyright whilst podcasts are out there in the open public space, if you want, and we can process that content easily by just respecting the copyright and playing back content giving the podcaster is more reach if you want. The situation is a little bit more complex when it comes to books, and you've probably seen that in the world of ai there's a set of lawsuits that are starting to happen left and center with regards. To copyright, both at the stage of, data used to train the models, but also at stages of how the content is being produced and how the data is used for, especially if the data is used for, financial gaining purposes or profit making purposes. So at the moment, it's something that we have in the back of our mind, but we haven't looked or researched into detail what sort of rights we need to have in place for us to be able to process and integrate books in our solution.

Dhaval:
Great. And with that we'll wrap up the show, but just one last question. You touched on something, you touched on a very important topic here, which is lawsuits and the source of data and privacy and copyrights. Right. So AI product creators need to be very mindful of the source of their data, the privacy and the copyright details of that data before they try to profit from it. What other advice do you have for AI product creators who are aspiring to build successful products in this space?

Calin Drimbau:
The most important thing is not to rush. , I think there's a certain frenzy in the market now given the abilities that the technology offers. So everybody's rushing in to, to build cool products. But in reality, for any founder, I think the important question is what is the quickest? What is the quickest route to value? So beyond and outside of building a cool in the product tool, what is it that you need to do to create value, either for a consumer or for business? And start from first principles of speaking to your customers, understanding a problem. Not start from the hype and not start necessarily from the technology itself. But once you figure out what problem you're tackling or how you want to use the new technology, then it's the right moment in time for you to accelerate. And, the APIs that are out there from large language model providers enable you to quickly experiment like never before. You can build solutions and products that used to take months, you can build them in days. So once you know what your direction is and what sort of business you wanna build, Iterate quickly, launch quickly, see what works. And then once you've landed on something that attracts the, user's attention and provides value to them then scale.

Dhaval:
Well, thank you so much, Calin. We really appreciate you joining this show, taking time to share your hard-earned knowledge and looking forward to have you back on the show in the future when you have more or lessons to share with us. Good luck with your pre-seed round. I'm gonna be following your journey and looking forward to connecting with you here in the near future soon.

Calin Drimbau:
Absolutely. Thank you so much for having me. It's been a real pleasure chatting to you thank you.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This award-winning founder built an AI Design Product and a user community of half a million users in less than 5 years
21 mars 2023· AI Product Creators
Tony Beltramelli is the Co-Founder & CEO of Uizard Technologies. Uizard Technologie is a startup developing AI-powered tools to transform the way people design and build software. He work at the intersection of machine learning, design, and software engineering. Tony Beltramelli studied at IT University of Copenhagen and ETH Zurich. In today’s episode, We explore the role of AI in modern product development, the significance of domain knowledge, and strategies for non-technical founders to break into the AI space. Tony shares the journey of Uizard, from its humble beginnings as an AI research project to an award-winning platform with a user community of over half a million. We dive into the world of AI and its impact on product. Tony also give advice for AI product creators. Tune in to hear Tony Beltramelli insights and experiences in building Uizard.

Find the full transcript at: https://www.aiproductcreators.com/

Where to find KD Deshpande:

• LinkedIn: https://www.linkedin.com/in/koustubhadeshpande

Where to find Dhaval:

• Twitter: https://twitter.com/DhavalBhatt

• Instagram: https://www.instagram.com/dhaval.bhatt/

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

In this episode, we cover:
00:00:00 - Introduction
00:01:23 - AI's impact on product creation
00:03:32 - The shift in the design process with AI
00:06:12 - Using GPT-3 for design and product creation
00:11:13 - The importance of domain knowledge and UX
00:12:52 - Breaking into AI product creation for non-technical people
00:14:41 - The role of distribution in product success
00:15:08 - Finding initial users and iterating on feedback
00:16:13 - Building a moat around your product
00:17:22 - Tony's vision for his product and empowering non-designers
00:19:09 - Advice for AI product creators
00:20:20 - Closing thoughts

Transcript:-

Dhaval:
This award-winning founder built an AI design product and a user community of half a million users in less than five years. In this episode, we chat about how you can gain a lasting advantage as an AI product creator. We also discuss product development philosophy for anyone who's interested in building on top of chatGPT 3. Tony Beltramelli is a founder of Wizard, spelled with a U instead of w. Uizard. Helps you build stunning mockups and prototypes in minutes.

Welcome to the show, Tony. Thank you for joining our call. Tell me about your product. Tell us about where you are, how long have you been doing it, et cetera, et cetera.

Tony:
Hey, Dhaval, nice to meet you and thanks for having me. Yeah, so I'm the CEO and co-founder of Uizard. So if you wanna look us up online, you need to look for Uizard spelled with a U instead of a W. and we are basically building an AI powered design tool to make it easier for anyone to basically build products, interfaces for mobile app web apps, you name it. Design is pretty hard. So we we bring AI at the core to just make it easier for everyone. .

Dhaval:
Yeah. Design is the last frontier, right? That's the part that takes a lot of creativity, a lot of lateral thinking. What was your process, thought process like when it came to building the product and how did you infuse AI in the capability?

Tony:
Yeah. So I've been basically like God brought into AI doing my grad studies. I did, I've been doing like a few projects back in the days. And then in 2017, back when I was working as a data scientist I was just still tinkering around with AI and deep learning in my weekends. And actually it's one of these like research project that was laying down the foundation for the company. So the company essentially started as an AI research project, before you even become a product in a company. So it was really like we didn't have to just back it up, at the end. It was just part of the foundation.

Dhaval:
That's awesome, man. What was the, when you found a company, wizard, when you define that company? When was it founded?

Tony:
Yeah, of course. So the, kind of like part-time, weekend project that I'm talking about was something I was building back in April, 2017. But it was just honestly like a side project and we only incorporated a company officially in 2018 with my co-founder, so early 2018. It took a long time to build the product, make sure it worked, make sure to solve a real problem, iterate around a customer, and then we launched out of private beta in February 2021. It took a while.Wow.

Dhaval:
Did you launch it on Product Hunt or did you have a list to go from

Tony:
we had both. We had gathered a waiting list of folks that had signed up to our, private beta and Alpha. But eventually, of course, when we were ready to go live, we, we also did a few launches on Product Hunt. We actually won, golden Kitty Awards for best AI and machine learning product in 2021, if I remember correctly. Oh wow.

Dhaval:
Congratulations. And how was the launch? What was the, can you share your, can you share about a little bit about the launch, whether it was sufficient enough for you to bypass the seed round or, yeah, if you can share any of that data.

Tony:
Yeah, of course. So at the time we launched, we already had raised roughly 3.6 million dollar of capital. So seed it took, it takes a lot of juice to just build accompanying product. I think Figma took, what, like four years to, to launch, which is like the same story with us. And so launch, it, it didn't like it never happened, like the the launch day and then. Skyrocket, Of course, it's just new spike of launch, you just nurture your user base and it takes time to ramp up. So yeah, it took a long time to just build the awareness build the right network effect in the product to incentivize folks, to invite other folks. But yeah, it took a while. Now we've raised, what, like more than 18 million US dollars. And we have, we are serving growing community of more than half a million users. But it, it took a while to get there. Wow.

Dhaval:
Yeah. So that's half a million users and the whole myth about all of a sudden you are founder of product market fit and you are getting pulled and your servers are crashing. That was like a romantic story that didn't really happen for you. You had to make incremental improvements that led to finding that fit eventually. Is that what I'm hearing? That's

Tony:
Absolutely correct. Throughout the. Two years of beta. There was a lot of product iteration. We had to just kill features, relaunch new features, test with customers. It takes a while before you can actually measure and quantify that you have product market fit. , it would've, yeah, that's the road we took. Let's just measure that we have product market fit before we put this live on the internet. But yeah, even though we measured, we had product market fit, it's still not an overnight success. Right. It took a while to just get to the first 10 K, 20 K, 50 k, and so on and so forth.

Dhaval: Yeah. That's interesting. There is a, it's a gradual process, right? It's not something that happens overnight and a lot of people have, all of a sudden you'll find the fit. .

Tony: No, completely. But then when it works, you can actually really see that it works. In the past, like six months, we've acquired and served more users than we've had in the, in the first year after the launch. it compound. And when the compounding effect works, you can definitely see it. It's a no-brainer.

Dhaval: Yeah. And the distribution and the right amount of, capabilities creates the pull in the market. So that's great. Thank you for sharing all of that. Let's dive into, Your, AI capabilities. Tell me how is it different than other design tools that are out there and, yeah.

Tony: Yeah, so our features are honestly quite unique. You can. You can of course assume that I'm just trying to market our own product here, but, you probably won't find this anywhere else on any other product. For example, we leverage AI to just enable our users to import a screenshot. So let's say that you are a product manager at Airbnb and you want to revamp that, onboarding flow when you get new people to sign up. If you were to just open Figma, sketch, any other tool, if you don't have anything already designed, you have to start from scratch. So what we've done is that we enable you to import a screenshot of anything, and then we use AI to just recreate what's in screenshot, so you can actually then go ahead and modify it to your liking. So you overcome the white blank page problem in just a few. Drag and drop your screenshot modified. There you go. You have your design. These other places where you, we use AI as well. For example. Let's assume that you are, like me, you know what you want to build, but you're not the greatest designer. So you can tell our ai hey, you know what? This is great, but it looks pretty bad. Can you please just copy and Paste the style of, I don't know, Twitter and make it look like Twitter? And so that's also a place where, our AI can just automatically do this, pull the style of anything and then apply it to your project. . Another example would be to, you, you can brainstorm ideas on paper or on the whiteboard with your team, and then you can just snap a picture of whatever you sketched. And then our AI will transform this automatically into a design that you can then modify. So it's really all these features are, is all about like, how can you, can we just simplify the ideation flow to make people. Focus on the core value, right? I'm trying to solve a problem with this design. I don't want to get lost into the weeds of like how many pixels to the right, how many pixels to the left? Should I move that button? It really doesn't matter when it comes to ideation, and that's kind of like what we're trying to do with ai.

Dhaval: Wow. I see you have a lot of differentiation within the product. One, one biggest challenge you are solving is the initial creativity block. That Some designers experience when they have a blank page and you saw that biggest problem, right? So now you have some something to play off of. And that's, that's awesome. When did you build the foundational AI model or ML model that serve the backbone of the product? Like how did you decide the use cases that, yeah, we are gonna solve for this use case and we are gonna use this AI capability. What was that process like? Tell me about the first use case. You decided to tackle with, ML?

Tony: The first use case was really like , the foundation project that I mentioned earlier that kind of gave birth to this company. And it was so just context on this when I was like my first tech job, when I was still an undergrad was front and developer. And I just hated the part where you need to take the design work and transform it into code with HTML and css. And so back at that point the idea was, hey, can we use AI to just take a picture of the design? And then turn it into code automatically. , the developer can just focus then on implementing the business logic. And of course, this feature was launched and eventually not used by our users, so we just killed it. But the core foundation of trying to understand the user interface, buttons, forms, what is an image, what is text field, what is a login screen? All this core foundation is the same that power or tech today. It's all about okay, we are gonna be operating within design, within user interface design. And if you . Teaching a machine, the basic concepts of that domain, then you can reuse it for all sorts of application. Like all the features are ultimately based off the same core whether it's for styling page generation, rename it, it's kind of the same core. so as a general principle, I would say if someone want to build a product for, I don't know, the automated industry, focus on what are the overall concept. Of the automative industry that you wanna work with, and if you can kind of like nail learning these concepts, then you can actually build multiple application layer on top without having to redo the core every single time.

Dhaval: So a new AI product creator or a product creator who's thinking of adding AI to her portfolio or existing product, the advice will be to learn the concepts, the industry, the inner workings of a domain. And then once you understand that you can turn that into a use case and add AI capabilities to those use cases very easily with today's technologies. Is that, am I hearing that right?

Tony: Pretty much think about, GPT 3 the reason why GPT three is just so amazing at, you know, being versatile. Is that ultimately they didn't try to teach GPT to, I don't know understand emails or understand customer support tickets. They focused on training a machine to understand English or language. And then once you kind of like nail that core foundation of understanding language, you can then do other things like customer support, email, copywriting. So like the same principle, like what are the building blocks of the domain you're trying to target and automate with ai, nail this. And then the application layer will almost, become, a no-brainer on top of that core model.

Dhaval: Would you say that with this new transformational technology that we have found ourselves being able to play with, would you say that for a product creator, domain knowledge has became. A lot more important now than it was before because a lot of generic stuff, all the typical product work that people used to do has been commoditized. And what remains is the product creator's ability to understand the domain and tease out the relevant use cases. Would you say that's a valid hypothesis?

Tony: I would actually agree. , and I would even say UX, the quality of the product around the AI is probably even more important than the core ai. I mean, if you think about chatGPT, which we just mentioned, OpenAI launched GPT3 in what, like 2020. There was like no traction. A bunch of, researchers, engineers were playing with it and had fun leveraging it for different products, but it's only when chatGPT came that, you know, in five days they had a million active users. But it's the same model. It's still GPT three under the hood. The reason why it's scaled so rapidly that the quality of the product. Leveraging this core AI was just mind-blowing. You could interface with it, chat with it as if it was like a human. And that makes the core difference. So as you said, like for new product creators, it's all about like the domain, the product that's gonna be, that people are ultimately gonna interact with. They're not gonna interact with the ai, they're gonna interact with the product. That's the ai. And if you nail this. Then you win basically, it doesn't matter , how smart or, it's secondary. the quality of the secondary, we even say,

Dhaval: Yeah. Let's change gears a little bit. Thank you for sharing that. One question I have for you is that you seem to be fairly technical co-founder. How would you recommend people who don't have that sophisticated technical background, what do you recommend them to break into the AI product creation space? Or is there a learning curve there? Can we use the existing LLMs, the large language models, their APIs or do you recommend fine tuning them? And then if there is process of fine tuning, what is the learning curve there? Like for non-technical person?

Tony: This is where probably gonna have, very different, opinion. Yeah. , but what I would say is that, Ultimately, it's very easy to build something around, on API, as you mentioned. Well, easy in terms of getting something up and running. It's really hard to build great product around it. , but so I, I guess the core, whether you are technical or not, is what's your differentiator that makes this product unique? Because they're gonna be like so many other products. And if yours was easy to build,then how do you defend against copycat? Right? You need to find a moat, whether it's, the quality of the product that's just so much better, whether it is a community of users. , so and ultimately the, the way companies win is not just by having great products, is also having great distribution and a way to just put this product in the hands of users. So, If you're a non-technical person, you might actually be really shining at the distribution part. And ultimately this could enable you to be successful, even if you can't, master the higher level of technicality that the your product offers. , so that would be one distribution ultimately decides who win and who doesn't. I would say

Dhaval: that's great. So distribution is one of the most underrated. Pm , product creation capabilities. And once you find the distribution, the important thing is to be able to iterate on the feedback you get from the users who are experiencing your product. For your situation, did you have a big team to support you with all of those iterations? How did you find that initial fit when you were going through those iterations?

Tony: well initially how did we find our first customer? How did we find our first users to get feedback from? Because we had basically built some very transformative tech. Just having a few demo videos online was actually enough to just initiate enough interest that some people will start signing up to a waitlist and from that waitlist, we could just then go and recruit people to just access the beta, the alpha and learn from their feedback. so I guess there is always like early adopters that are looking for trying new novel tech. , and so AI has this just magical effect that people certainly wanna try, right? So you can totally leverage this to your advantage and collect, signup, email and eventually get the first 10 or 100 people to try your product and iterate from.

Dhaval: Yeah. That's awesome. In your case, what was that moat when, earlier you mentioned that you had to find a moat, whether it's a technical advantage, user experience, community or distribution. In your specific situation, did you find that mode, through your product, through your distribution? What was that like?

Tony:
Well, I guess the moat is something that we are constantly building, building and refining, right? Because as a company grow there, the moat changes over time. Initially it might be the. The deep knowledge that the founding team has for a specific problem that no one else understand. Like this is your unique selling point that no one else has and can replicate, but eventually the quality of your tech, quality of your product, quality of your community. But , early on we had no moat I mean, when we were, first building this company, if anyone had the same core interests about a specific problem, we're trying to solve it would've been fairly easy for someone to replicate it , if they started at the same point in time. So early on, I would say your moat is like how deeply you understand the problem you're trying to solve. that makes it hard for anybody else to just, solve as well as you would but over time, for us it became clearly the product and the community. We've had really put an emphasis on community led growth which ultimately it's hard for other companies to replicate. plus the core technologies, which is still not available in any other design tool. , we still are the only one with the features that, I mentioned earlier.

Dhaval: Yeah, so it looks like you found moat in two places. One was in the community led growth, and the second was in your core technology advantage of understanding the problem and identifying the most foundational use cases like initial writers or initial creative blog. Looks like you really have an amazing roadmap ahead of yourself tell us a little bit about where does, where is your vision? What is your vision of the product?

Tony: Yeah. The vision of the product is really to enable anyone within the product team to take an active part in design processes. Design is increasingly important. You can probably, like, every product you use, you will immediately judge this product based on the quality of the design and the UX of this product. You might actually just not even look twice at that new app you just install if it doesn't feel good to use. But the challenge is One Figma is really hard if you are not the designer. The number of time I meet a manager or team lead that doesn't know how to get around Figma is just insane. And two, even the few folks that understand how to use Figma well, they're already pretty busywith all sorts of design work and product discovery. So our vision. How can we just empower all the other people in the team that don't have the time or the skills to use Figma to still take an active part in product design? Because chances are is not just the designers or the product team that have good ideas. The customer success are team is hearing feedback from customers every day. They have some kind of nutrition on how to fix it. And so if we provide them a tool that enabled them to. Materialize the vision that then ultimately they can help the product team, move faster towards solution and everybody wins. So core vision is again, like put design in the hands of as many people as possible so companies can build better product faster.

Dhaval: Thank you for sharing that and we'll end on this question. What advice do you have for product creators in new to the AI space or existing product creators in the AI space who are trying to excel

Tony: it might not be the best thing for a lot of folks to hear, but you need to get deep into the technical details. You can build a lot of things around API, but ultimately you don't wanna build a product that would just die if OpenAI decide to just turn off the api. And so if you wanna build something at last, you really need to understand what's going on under the hood. And so I would say, Get at least an intuition of what's going on. So you understand at least you know how to move things, ar how to move forward how to recruit the best engineers and how to have more proactive process around improving your product. A great place to start would be the FAST AI course, which is free maintained by Jeremy Howards which is an amazing place to just you. Get into ai really quickly and for free. So go ahead and, and learn the deep stuff.

Dhaval: Great. Thank you so much for being on the show, Tony. We learned a lot from you. Thank you once again.

Tony: Thanks for having me dhaval
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This founder built an AI Content Workflow product and grew it from 0 to 1M users in 1 Year.
20 mars 2023· AI Product Creators
KD Deshpande is the Founder & CEO at Simplified He is the entrepreneurial product leader who founded two SaaS companies, built teams & products from grounds up, raised venture capital, and successfully sold the business through an M&A exit. Marketo acquired his last company Vessel.io. In today’s episode, KD talks about how he bootstrapped his idea for the first few months and grew his product using the minimum product payable framework, which is the term that he coined, and his product development approach using OpenAI. KD also shared valuable insights on how enterprise product creators can find the sweet spot in their products and how they can leverage AI to create personalized and defensible products. KD also shared some tips on how to attract and hire the right talent for your team. Tune in to learn more about Simplified's approach to create personalized and defensible products, and some tips on how to attract and hire the right talent for your team.

Where to find KD Deshpande:

• LinkedIn: https://www.linkedin.com/in/koustubhadeshpande

Where to find Dhaval:

• Twitter: https://twitter.com/DhavalBhatt

• Instagram: https://www.instagram.com/dhaval.bhatt/

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-
Dhaval:
This founder built an AI product to replace your entire content creation workflow, including writing, design, videos, animations, and social content. KD Deshpande is a founder of Simplified A Product. He grew from zero to a million users in one year. and has continued on that growth trajectory since then.

In this episode, we discuss how KD bootstrapped his idea for the first few months, and specifically how he grew his product using the minimum product payable framework, the term that he coined, and his product development approach using OpenAI. We also discussed how he recruited his initial founding team and his approach towards recruiting early-stage founder.

Welcome to the show, KD. Thank you so much for joining us. Tell us a little bit about you and your AI product.

KD Deshpande:
Hey, thanks for having me here. I'm KD founder and CEO of simplified. Simplified is a one app for all types of content creation. It's a modern age tool where you can do more with less. We are, as a consumer, we are consuming more content then ever it's a Cambrian explosion of content creation and the tools, existing tools are slow, siloed, and we are fixing that. We are bringing the operational efficiency and truly simplifying the entire workflow from creation to distribution for creators like you who are putting a lot of efforts in creating content, inviting us to be on your podcast, but we want to make sure that we bring the efficiency in your workflow. In every content creator's workflow, that modern marketers workflow, so that way you can do more with less.

Dhaval:
Tell us a little bit about your market. Who do you serve? Is it sales? Is it marketing? Is it content teams in product? What is your ideal customer segment?

KD Deshpande:
That's a great question. if you look at like last, probably let's take the last four years. The bar for content creation is lower that with Instagram reels, Facebook, Twitter TikTok especially, they reduce the barrier to entry for anybody can be creator, anybody can be marketers. So if you look at our traditional influencers or marketers are looking like influencers, and influencers are looking like marketers. So, anybody who creates content, Anybody who help companies, businesses, to market their product, we serve them. But primarily, we help digital marketers, digital creators online, small businesses who are focused on creating content. Because content is no longer liability, content is more seen as investment because the brand, the vicinity or the people are shifting from brand vicinity to like influencer affinity. So that's kind of the target audience we are going after. Yeah.

Dhaval:
You mentioned something very interesting here. People are moving from brand vicinity to influencer affinity. People relate and have always related with human beings more than with brands and very interesting direction here. How does it, how does your product simplified? How does it help creators? specifically, what does it do for them? That resolves their pain points. And where what role does AI play in this?

KD Deshpande:
Yeah, so AI is really amazing, and last two years we have made tremendous progress, but we not officially saying, oh, we are doing ai. It's part of our DNA our AI is like Gmail Smart Compose it's there. It makes you smart, but Google never says that we are doing that by ai. So that's what simplified ai is. What we are doing is we are bringing the best technologies off the shelf as well as we are building in-house. And you, we are humanizing those. We are humanizing to a point where That marketers sitting in Utah or a marketer running a digital agency or a creator, running creator who is, like recording their next TikTok sitting in Brazil or India. All of them can use simplified to reduce, their time for creation. And then we help them optimize their workflows. Where on simplified it's a one app. Because if you remember, if you look at their workflow, probably your workflow, you have like probably five apps before your content gets ready. From step one app is just for recording other is for curation third is for publishing. Fourth is for analytics. With simplified, we, we are bringing all that workflow in under one umbrella. So that way you can reduce, we can reduce your cost We can save time, so that way you can spend that time doing something meaningful with your friends and family and save a lot more money for these businesses.

Dhaval:
Wow. So, that's so true. I use five different apps for creating anything I create. There is Like a place for me to research, then there's a place for me to curate. There is a place for me to distill. Then there is a place for me to create once I have distilled and then there's a space for me to edit and publish. And then there is a space for me to analyze. So seven or eight different steps. And each step has its own set of tools. Yep. , and you are saying that you are aggregating all of. Stack into one capability.

KD Deshpande:
Yeah. We are aggregating more than ag aggregating this stack. We are trying to simplify your workflow. Everything you need as a creator, as a marketer in your workflow. We are bringing them making it available as a part of Simplified Stack. So you just come in today, you can come in, create your you can start with ideation. So we have ai powered like GPT3 powered Writer. You can come in, and we have done a lot of fine-tuning modeling on top of it and built a really simple user experience because see as I said, that marketer or that small business owner sitting in Utah or like other places, they don't care about, GPT 3 three or ChatGPT or Stability or Dali-E. Wh what all things they need is like simple. Which is, which allows them to do more with less. So we have AI writer where which can help you create a lot of ideas. Then we have design editor which can turn your ideas in one, click those ideas into presentations, Gif memes, videos, we have templates. Then you can turn that, take that once that content is ready, plan for month or so you can start scheduling and start publishing on all the channels from Facebook, Twitter, LinkedIn, TikTok, Instagram your WordPress. From one place, you can just go from creation to delegation to distribution. All happens in one platform. So that's what the problem space we are dealing with. And under the hood, we are integrating all the AI services. So that way they are integrated into your workflow. So that way while you are doing it, the things which used to take you probably two hours, 10 hours, we are trying to reduce that in minutes, using the modern age technologies and the capabilities which we have seen, the extreme revolution in last 12 to 18 months.

Dhaval: Yeah. Tell me a little bit about how is, how are you differentiating? There is a plethora. AI tools that have come out in the recent history, how are you differentiating yourself from that competition as a product owner?

KD Deshpande: Yeah. to be, to give you a fact, there are probably 85 to probably 90 AI writing tools. But all these tools are build silo and where we are thinking differently is. Our tools or our stuff is all about workflow. We are thinking that persona about, of that marketer, how that creator we are living the life of that creator ourself me I'm going recording TikTok reels, trying to live and my co-founder also going through that journey. Our designers on our team are experiencing what it takes to create content. what that life of that small business owner looks like or how the modern marketing teams are working. And the, that's how we are fitting AI into their workflow instead of adding one more silo tool in their stack where they will need to copy paste this from, someplace to their day-to-day workflow. So that's the big differentiator. And second is our product is meant for teams. our whole purpose is let people collaborate. And our mission is not a build a product we are building a space where people can collaborate together , do more with less, and unleash their creativity. And creativity can come in any form. It could be ideas, audio, video, all those stuff. So that's the biggest differentiator between existing things out there and what we are building.

Dhaval: Wow. So you are starting with the workflow first mindset. You're taking the existing behaviors existing workflows. You are bringing that into your product. And the second thing I heard is that you are creating a space for people to collaborate and work together. And those are two of your foundational differentiation compared to other siloed work tools that use ai. Yes. Tell us a little bit about your company's State right now. Have you bootstrapped? Are you raising capital? Have you raised capital? Yeah. Give us a little bit of a history on how did you start, did you have funding? Did you raise seed round and where are you going?

KD Deshpande: Yeah, so I started this company at the start of the pandemic when I was, so my journey is, this is my third company. I've been in marketing in and off. I'm an engineering entrepreneur. The started my career as engineer, then became entrepreneur, sold my last company in 2015, and then worked at Uber, Facebook, and I was, at Facebook. When I joined Facebook in Feb, I was moonlighting on this idea for a few. And then I convinced that there is I've seen small business owners, I have seen Fortune 500 companies, and there are amazing marketing automation tools available. But when it comes to content creation, small companies don't have resources or small business owners and big companies have resources, but it lacks efficiency. So to solve that problem at the start of pandemic, I started this company, first six months bootstrapped it, and then raised funding from like several TRAVCs in the Silicon Valley. So we are two years in business. We all, last time when we officially announced we have 1 million plus users, 1 million plus in a r r. Um, and right now we are just focused on like really happy community of paid customer. We are just focused on building this ecosystem of product and keeping up with the pace and making sure that, let's bring that simplicity in the creation process, bring that efficiency with AI in this process.

Dhaval: Got it. So you have over a million users. Your ARR is over a million dollars and you raised seed round and you haven't raised anything since then, are you. Are you at the point where you don't need to raise capital for the near future, where you are just riding the wave of success you have created? Or are you getting ready for that next big expansion?

KD Deshpande: We are we are capital efficient because we are fully remote team we have like different hubs. That's how we are capital efficient, and we are riding on the success the way we, in a short period of time. We, last year we grew from zero to 1 million plus in a short period of time. So we are continuing on that and since we are capital efficient, we don't, I mean, as a founder and CEO you never say you are not raising. But we have enough runway to continue, on this trajectory of growth. We are on already

Dhaval: tell me a little bit about how did you manage to bootstrap yourself for the first six months? Was that off your previous successes? Big tech, savings, or was that from some other innovative method?

KD Deshpande:
Actually that's a very interesting question that, I wish somebody asked me this earlier. It's a combination of both. I even though I transition into product management, I still like to code. It keeps me sharp, and it I like to, I always say professional are as an engineer or as a professional knowledge workers are, we are very identical to. Players in games. we have to keep ourself up to date. So I've been keeping up to date. I've been keeping myself up to date with like new technologies, so I started coding by myself. First, I started doing a market research. I posted funny fun fact. I posted few jobs on Upwork and Fiver just to interview people saying, Hey, if you use these design tools, here is $25 gift card. I would just like to see how you design I would like to understand your processes because building design, product or creativity tool is it's not a job of lighthearted people because, people expect, your tool to perform at certain level. So I first six months, my job was to every Friday just schedule the calls with this freelancers, designers and get on a zoom like ourself and watch them designing. And then I was coming back. I hired a couple of , engineers from my, previous jobs and we we built a prototype. So every Friday, and since then we have Friday show and Tell. So I was showing the product, they were thrashing our product left and right. We were coming back, taking that feedback, fixing those bugs, or creating our roadmap. And then next Friday, again getting for that conversation so we know that what the last time the, with the new set of users this time, so we knew that, okay, this users faced this problem. We already fixed that. So we did that, this continuous discovery. And that's how we kind of got closer to what people are looking for. And as, as compared to my last companies or the product I have built in different places, we no longer in a phase of building minimal viable product. I think it's a old term. We are in a phase of building minimal payable product and when we there, there is a very fundamental difference. I coined this term minimal payable product. When you say minimal viable product, all your friends say, yeah, I'll use it, it's good. But when you say, this is minimal payable product, they have very different cheat sheet, table stake features. They would love to see before they can make a switch. And our goal was let's build this minimal payable product. So before we publicly released, we knew that that product is ready where somebody can switch even in the crowded market, we can make our own stand. And that's how we are in spite of like, as you said, in spite of crowded marketing design space or like the really amazing what Canva built in this space or last decade, we'd still be able to make a market. We are building brick by brick and that DNA is still continuing in our product development, which is go up to our customers, ship every Friday show until, get more feedback, come back. I trade products fast because as startup. Moving fast and that's your advantage. And that's what we are doing right now.

Dhaval:
I love the culture you have created and I love the term that you just mentioned, which is minimum payable product. I think you need to talk more about it I think more entrepreneurs need to embrace this minimum payable mindset beacuse creators don't think of the revenue side of businessmen they're creating. And the difference between a creator and the entrepreneur is this specific thing that you brought up, which is minimum viable product versus minimum payable product. Thank you for sharing that.

KD Deshpande:
When we were, if you go 10 years back, You, the bar as a user, your bar for a product was probably low. You are still okay with like non-collaborative product which you can use that does this one thing and does it really well. But when you go in the market right now, you expect it, it to have like sign in with Google. You expect it to have some collaboration, invitation workflows. You expect it to work flawlessly and accept credit cards that's where this minimal payable thing comes in, where you should be able to say, okay, it looks good. Here's my card and start charging me. I don't mind it switching from any existing product. So all fellow entrepreneurs or the builders who are building right now, my general one piece of advice is just go ask your people like, what is minimal payable product looks like in your own category? Why at what point someone will. Switch and say I'm fine switching from this X app, which I've been using in my day-to-day workflow to your app, and I'll pay until you don't have that. I don't think it's a good idea to put your product out. You can still do this like I treat you development until you fit that.

Dhaval:
that's one of the most important product management lessons for early stage entrepreneurs that I just heard. We'll. We'll make sure that we unpack that a little more. How does this apply to enterprise product creators who are working in the in the larger space where they, they have the luxury of investments and the backings from the corporate sponsors. Where do they find this sweetspot

KD Deshpande:
I think they, they have probably, they have biggest advantage than anybody else because you have, like quarterly business reviews with your top paid customers. So usually you can get them in the room. When I was at Marketo, we used to use, used to have this quarterly business meetings with all the top 25 customers or a product council where you can go and bounce a lot of these ideas. and just , understanding the persona, understanding their workflows can give you like significant advantage in your own category. Even you are established like established product out there. And of course, like in the enterprise space, the pace of innovation is still slow. But still, if you find that those champion customer. You partner with them more frequently and see the usage? I think the, once you put things out there, tracking more, having those KPI based culture usually helps in any sort of the company which is out there. Just understanding like, okay, how can I grow business? Because when you are in a, when you're a startup, it's about like, can someone pay me X dollars when you are an enterprise? It's can they, can I upsell them? Can I cross-sell them? Can. What can I do to increase my ARR or like ACV, for that customer? So mindset is different, but you have a lot of resources. You can even have like analyst to go and do these, research for you and get back, like understanding the market, what's happening, keeping, keeping, an eye on the new technologies like, like chat GPT or stability, ai, all these new trends because. Waves don't happen that often. And if you don't and a AI is going to disrupt the way we do used to do business. So the companies which are not going to ride this wave, I think there will always be new players coming in and replacing the incumbents.

Dhaval:
Thank you, KD that was really valuable insight for both spectrums of PMs, right? Product creators. One, one follow up question. You brought up the new writing wave of writing, the wave of new technology let me dive there a little bit. You said that you are technical, somewhat technical. How was your decision around building on top of. ChatGPT or OpenAI, APIs how did you make those decisions? Did you build your own models? Did you fine tune the existing models from those other large language model companies? And where did you draw the line? Yeah, this is how much we are gonna use pre-baked and this is what we are gonna bake ourselves. What was that thinking process and how should others approach this decision to make a defensible product?

KD Deshpande:
I think, let me unpack this question because it is a really important question and it's important too. So first thing is like, how did we start? So if you look at it, anybody we've been hearing AI is going to replace you. AI is there for the last probably 20 years or like since you started into this engineering field, but up till now probably. I would say OpenAI is probably the major breakthrough up till now. This technology was available of two big companies with lot of resources. And this probably first, first time ever it may OpenAI, made it accessible to every single developer out there without worrying. About GPU resources without worrying about like heavy lifting to training those models because these things were not easily accessible to average. Product person, average engineer, I think OpenAI did that first, that breakthrough to kudos to their team. So when we started again, we looked at the workflow. If you look at design most of the designers. And the workflow marketers want to create this amazing marketing assets. They ask designers, they brief designers, and designers go and come back with some placeholders like Laura mercier some placeholders, and then there is a back and forth. So we said, okay, what if designers have this superpower? They know the brief and they can just put into this like magical tool magic parks, and they get this. Like probably some input box or like some catch catchy taglines. Alright. Ads briefs. So we started looking into this. We looked at, some of the Stanford models, like some of the G P T models. And then eventually when we stumbled upon, or GPT GPT 3 at the beginning of when open AI was just experimenting, we did some prototype first versions were without any fine tuning, without any modeling. It did this, it did okay, but then we quickly learned that these out-of-box models are good. But it's almost like a, I always say to, in my sales calls, AI is almost like a smartest intern in your office. They don't know what to do, but they can do a lot more if you instruct them really let's sync this start, sync in this start so that in turn, if you guide them, they will do really well. So we had to write that programming layer on top of it for cleaning the data fine tuning the data so that way we can, because our use case was very design marketing oriented. So when marketer is using the product or that creator is using the product. They expect some sort of language, professional tone like depending on their campaign, depending on their brand presets. So that's where that layer on top of came in. And then we, as we matured, as we started building the ML bench strength, we started running some panel models to enhance, these things. So we, we do build versus build like a hybrid approach, right now because these companies are pouring millions of dollars to run this infrastructure. And we, what we are doing is we are what some of the things we are building in house and some of these things, we are using it off the shelf. So that way and on top we are building a lot of data cleansing, training, fine tuning, so that way we can build defensible market. And third question is, how can in , this market, how can people. Build defensible product. I think future of AI is inter personalization. I think that's where, take that use case and personalize lot more stuff and your AI should get smarter as people use it more. I think that's the, it's a constant process. It's not like you can do it like one time and you are done. it's a constant process. You need to like refine your AI offer unique solutions specific to that particular, problem and define clear ROI. Otherwise, it can become a research project very quickly with like massive GPU bills. And you can burn through your cash reserves in no time before you can humanize or productize that use case.

Dhaval:
Thank you. Thank you for sharing that. and in your specific situation you mentioned that you started off by using the existing ML models, as your initial prototype, but then you invested heavily in personalizing and fine tuning this. Where did you find the talent for this? Where should a non-technical product owner go to look for people who can help with building this fine tuned, personalized experiences for their users?

KD Deshpande:
Luckily that's we've been building our product in public. We've been sharing a lot more details luckily a lot of good people once you build a good product and build a good culture, you attract that kind of talent. So our ML team is very organically built through references through like through communities through Twitter dms. Our last ML engineer we hired through a Twitter dm. The latest one someone who joined. We are building a remote company. So our, one of the engineer you won't believe it he's joining now from Morocco and specializes into ML who is joining our team to focus on all these different capabilities because some of the things. Which we are building, that, that are core to our business. We are building those in-house. So we are, we have this staff and we have a very, we acquired some of these through acquisitions as well. So we have a very strong team with the right skill set and right experience. And hopefully, I mean, not, hopefully we are in a right domain at the right time with the right team. That's the kind of beauty of where we are right now and yeah, that's like you, you just create good culture, put it out there. Join those communities. There are good communities around Reddit, Quora see what are the people talking on Twitter? You follow up with them, you build the relationship and then slowly you start bringing them because that talent attracts talent. So that way, you can create the culture that provides opportunities for people, and that's simplified. We have very simple principles people are P zero that means treat others the way you want to be treated show, don't tell. You may be like, great a scientist, but if you can't demonstrate your skills probably you might, simplified, might not be the right place for you. So second is show and tell, keep things simple. We want to help our community. We want to help our customer. In a way that we are simplifying their workflow. So keeping it simple, the way we raise money, the way we do business, way we do partnerships, it's all part of that, these three philosophies. So that's how we are building, attracting, and hiring, recruiting talent.

Dhaval:
Thank you so much KD that was very helpful. And that concludes our show. I really appreciate you hopping on the call and sharing. Hard earned knowledge with the community and looking forward to have you back when you have more success stories to share with us.

KD Deshpande:
Absolutely. Thank you.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This Founder Harnessed AI Language Models in Just 2 Years, Revolutionizing Copywriting with 850,000 Users and Exponential Growth
17 mars 2023· AI Product Creators
Yaniv Makover is the Co-founder & CEO at Anyword. He has done research in the fields of Machine Learning and Natural Language Processing. Yaniv also served as a lieutenant in the Israeli Defense Forces. Anyword's AI Copywriting Platform and also the world’s first Language Optimization Platform that helps publishers and growth marketers deliver and optimize the messages they use to deliver business results across web, social, email, and ads. In today’s episode, how Anyword leverages large language models like GPT-3 to personalize copy for different segments of a business's audience, providing insights and analytics on how well it will work for specific target audiences. He also highlights the importance of prompt engineering and the value of feedback loops to improve copy performance. He discusses the challenges of building an AI product, emphasizing the importance of staying focused on specific problems and being disciplined in product management to ensure the best user experience. Tune in to learn more about Anyword's approach to AI copywriting and the future of personalized copy for readers.

Where to find Yaniv Makover:

• LinkedIn: https://www.linkedin.com/in/yaniv-makover-a8590b3/

Where to find Dhaval:

• Twitter: https://twitter.com/DhavalBhatt

• Instagram: https://www.instagram.com/dhaval.bhatt/

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:

This founder built an AI copywriting product and got over a hundred thousand users in two years. In this episode, we talk about his approach to deciding where to invest your time, money, and energy, as an early stage AI founder and I learn a lot about his approach towards prompt engineering. Yaniv is the CEO and co-founder of Anyword leading AI copywriting solution designed for marketing performance. He has a Master's in Computer Science and Information Systems, and he has conducted extensive research in the fields of machine learning and natural language processing. His work optimizing ad and content channels for some of the largest publishers like New York Times, Lead him to found Anyword generative AI platform. Yaniv oversees operations across Anyword’s New York, Aviv and satellite offices across the world.

Dhaval:

Welcome to the show. Yaniv. Tell us about your product.

Yaniv Makover:

Hi. Thank you for having me. Yeah, I'm co-founder & CEO of Anyword. We are in the coparating AI space. Our product primarily focus design on making copy more effective and converting more and engaging more.

Dhaval:

Wow. Okay. So Anyword and. how is it different than the plethora of other copywriting tools, AI-enhanced copywriting tools that are out there, at this time, or they're probably gonna come out now that there's a lot of those capabilities? Yeah. Tell me about how is it different. What's the differentiation there?

Yaniv Makover:

So we started out from becoming performance mindset. And I think there's one. Big problem that generative models can solve, or language, large language models can solve is basically helping you just get more content out there and high quality content removing solving for writer's block. And I think that's an interesting, huge problem to solve. For us, it's kinda like not our DNA. Our DNA is more about in our products DNA is how to make your. Copy better. So you already are a marketer. You already know what you're doing. You have a strategy. Yes, you could get ideas from ai, but this is how Anyword will work better for you for a specific audience. Or you're selling, I don't know, a sweater to somebody in the US versus other countries or a different occupation or different age or different gender. Then how, what are the best words to use for every use case and I think we use large language models to actually empower those insights or actually leverage those insights, to create ROI for our customers. I also think that when you're just a click of a button away from creating hundreds of variations of copy we thought it was a really big problem to solve. Which one are you, you're gonna publish, you can't really A/B test 1000 tweets you have to send one. And so we thought that was like the biggest problem solved. So we, early on, we focused on that. And our product, we pretty much tell you if with every copy variation that the AI generates, how will it will work and for whom and why. And if you wanna make improvements, how to do them.

Dhaval:

Cool. Okay. So you're taking the existing. LLMs and you are not only generating content, that's an easy problem to solve, but what you're really doing is you are helping fine-tune that content to resonate with the specific audience that your customers may have, and then fine tune the copy to increase engagement or retention with that audience is what is the primary metric that you aim for with your customers to improve? What is their North Star metric that you're helping them?

Yaniv Makover:

So typically it depends on the use cases of the marketing, but they'll, they'll measure lift in conversion rate or lift in engagement, and then they'll measure just ROI so if they're running ads, they'll, they should be able to see a lift in their ROI or if they're, the conversion rate on the lending page or open rates and emails. And it's pretty easy to measure easily. Just copy and see if see if it works for you.

Dhaval:

So do you, how do you get the information on their audience, like to fine tune the output with that highly fine-tuned output? Yeah,

Yaniv Makover:

so Anyword collected its own data, and basically, we have our pretty large corpus of data, performance data. And also when we, we partner with our customers, our partners, basically they have their own data sets, and then they upload in them into Anyword, and then we have, we fine-tune what we call custom models to help them predict better how their copy will do. So, for instance, just based on our data, we have an accuracy measurement of how well our model predicts performance of like copy that we already knew how well work. Somewhere around 76% depending on, on what we're testing. But if you, bring in your own data and you've actually A/B tested or just ran a copy in the past, then it goes up to 85. And that's just because you have your own audience, kinda like your own topics. I think for me, one of the most interesting parts of the space of large language models, not only they can write really well, they also understand text really well. So like five years ago, if you train a. Just lots of text and tell it, this text is good, this text is bad. We'll probably figure out that one has an emoji or an exclamation mark. But now there's a deep understanding of why text works. Like are you using, if you're missing out is that even relevant for some audiences and for some products or industries? And I think that's super exciting. So I think it wasn't possible a few years ago. And it's possible now. And I see this as kind like a. Booster and performance for marketing.

Dhaval:

Hundred percent. There has been a plethora of content generators using chatGPT and tools like that to create content.

Dhaval:

But what is still missing is the ability to fine tune that output to the specific segment of your audience and then be able to create content based on that readily, readily, as in with a click off a button. I'm sure you can stitch together a few data pipelines. Do that with existing tool suites. But what I don't see happening is the ability to just readily click of a button, say, this is my audience, this is my content. Generate some copies. Is that, am I understanding your pro product correctly? You offer that?

Yaniv Makover:

Yes. For every, like, while you're typing, you can even not, you can use your own copy, not even generate with ai. You'll have insights, analytics about that. Copy how well we'll do for your target. The way you defined it, what talking points work better and maybe replace them with others? And you can use your talking points that work for you while you're running ads in your emails and the talking points that work in your emails or the insights you gain from that and your landing page. And I feel like that is kind the future where there's just so much content that's gonna come out. You really know you have to know what, what resonates with your audience.

Dhaval:

Yeah, I imagine a world where the specific copy will be highly personalized to the reader that is consuming that information. And it could be hundreds and millions of variations of that depending on who's the reader, right? So what I am curious about is when did you. Now that I have the context of your product, let's talk a little bit about your business. Tell me about when you founded the company, and, tell me a little bit about the number of users you have amount of your revenue, et cetera. Anything you can share to provide us context on your business?

Yaniv Makover:

So Anyword launched March 21. And basically, it was a spinoff of the first company we founded, which called QE QE it's a SaaS platform for publishers, New York Times, CNN Washington Post helps them. Distribute their articles and their content on, on, on social platforms like Facebook. And then what we figured out there and QE works with 70% of the top media companies in the US What we saw is that some of our customers did way better than others in engagement and conversion rate, just based on their copy. And we thought, okay, we can help our other customers to, to write better copy. And then that was kind of the. What made us think about Anyword? And then the problem was much bigger. Not just copy for social posts, but you can even rewrite today a whole article for five different audiences. So you might be reading a different version of that same article, then me and the author. It's the same the same author with the same idea or message. But we'll be reading different articles that talk to us based on our. Words and familiarity and language. And I feel like that's super cool. I think I always thought that was like, a future and I think it's very possible today. I think something that, pretty profound.

Dhaval:

Yeah. Yeah. That, I think that's pretty cool, actually, to be able to have such a level of personalization for all the segments of your audience. What I'm curious about is, Do you have, is this product out already? Do you have any customers using? I know your previous product was mega successful. 70% of the media companies using a product. That's amazing. Congrats on that. Success, QE QE, right? Now with this that you launched in March, 2021, Anyword that was launched in March 2021, what are some of the business results that you could share with us.

Yaniv Makover:

So we have almost a hundred thousand, users using the product. And we have a few thousand paying customers. And we've the product is like the our top line is is now at an like, exponential rate growth. So chatGPT was a big help Introducing the whole economy, this space. And I think, as far as product market fit, I think there's a hunger for helping, in copywriting, specifically for performance copywriting. And super happy to be in this space and yeah, to grow with it.

Dhaval:

Yeah. How many to? What's your pricing point? Are you are you enterprise? Do you have enterprise customers, individual copywriters, or freelancers? What is the market segment that you aim to grow and expand?

Yaniv Makover:

So we have the comic starter offering. It starts at $29 a month for consumers pursuers. And then, like half our revenue comes from enterprise, where we are API powers ad vendors, agencies, big brands that have a lot of data and, trying to leverage that data to create performance. So I wouldn't say their biggest issue is to create more content or copy. Their biggest issue is what we're basically helping them do is this is the easiest lift they have in their, their easiest 10 or 15% lift in their marketing funnel that they'll get just by improving the copy. I feel like it's a must happen. I think every marketing org will have a box like this in their stack, to improve their copy.

Dhaval:

Hundred percent. Yeah. It's gonna be part, it's gonna be part of every single stack, all the way from creation to publishing. Right. I totally see it like becoming an ingrained part of copy creation and copy publishing and all of that. Yeah, now tell me a little bit about how you stumbled on the product idea. I know you. Talking about how you had media company and a company that was used by a lot of a product that was used by a lot of media companies, 70% of the media companies, and they had challenges with the conversion rate on you noticed that you could improve that? Is that how? Tell us a little bit about your origin story and tell me also about whether you are a technical founder if you have another technical co-founder.

Yaniv Makover:

Yeah, so I'm a technical founder, and I also did my master's degree in national Honors processing. And my co-founder is also technical. So we're both of us are technical in that aspect then, and even before large language models like that makes I really believed in in, in our ability to improve performance of copy. So when I saw that, some publishers were doing better than others and not necessarily the ones that have. Had a bigger budget or bigger teams or adjusting what they were doing, and they were A/B testing more copy they're doing way better as far as engagement and conversion subscriptions and the, and sales than the other ones. I thought this was, like, something we can help with AI early on we are just struggling to take like an existing social post or a tweet from a publisher and just replace two. To have it make sense and perform better. And as models became larger and foundational models became larger, we could leverage that to solve all those contexts issues and, that kind of really played in our favor.

Dhaval:

Wonderful. Tell me about how you created the product. What is your stack like? What how long it took you to create the product? What is the size of your engineering team? The reason I'm asking that is because our audience are people who are either interested in creating an AI product or have. How to Infuse AI in their existing stack. They work for an enterprise, and they want to do that. So give us an estimate of like the budget or rather, the size of your team and the effort it took you to build this product. Yeah.

Yaniv Makover:

So there's around 25 people in the economic engineering org. I think there's a big difference between leveraging. A foundational language model likes GPT 3 or building and fine tuning your own models. So which like GPT 3 or others like it, you can build really quickly. Like it's,you don't need data no need for fine tuning and you can just focus on workflows and user experience. If you have your own model, then it's way slower. You have to there's a process for obviously training it, but also, you know, getting to the right performance, and that could take months. So I think when you, that's like a big decision. We invested early on in a big, like a, in comparison to our team, a large AI team. So there's five people in the AI team, and they, we train models, not not only. Write prompts. So prompts are much easier to write. And I think about decisions like where you need to invest in your own models and when you don't. I think you can iterate really fast on user experience, product market fit and get initial solutions going with foundational models. And you really need to choose your battles where you absolutely need, if you actually need one your own fine tune model when you want to host it, because I think that's Much bigger investment. What we found to what worked for us is yeah, you can iterate really fast with just prompt engineering and see and get to initial result and then go deep, so to speak is, to build like your own models to, to solve some specific problem, for instance, and when you have to predict what copy will work better for others or if it's more. in line with a tone of voice that the customer's looking for, then we had to really fine tune our own models and that's an investment but you, when you just want to create a new format for framework for copy, then yeah, just do it with a prompt and you're good to go.

Dhaval:

Yeah. So for a product creator who's interested in creating an AI product, there are three areas. She could be investing in. One is learning the foundational models and iterating use cases and workflows. On top of that, essentially a UX and a distribution advantage. This is the first use case. The second use case is the building, fine-tuning and personalizing the foundational model itself. And that could be a lot of investment. That's not just a distribution or a UX advantage. That's a technical defensibility of the product itself by iterating and improving the foundational model itself. And then the third, the found, and then the third area that she could invest in is prompt engineering on top of fine-tune models or existing LLMs. And that itself could be that's a, that's an advantage in terms of problem discovery.

Dhaval:

How did you improve your prompt engineering skills? Looks like you fine-tuned it to a point where you have hundreds of thousands of users. What advice do you have for someone who is trying to focus on that category of improving their prompts?

Yaniv Makover:

We invested in a feedback loop really early on. So we actually connect to marketing channels. Our customers marketing channels are Google ads, Facebook ads, API their website. And then we really care about performance. If you know what you're looking for, you can measure and then and then figure out what's wrong with your prompts. If they're not creating the right outputs. We also have these. Quality surveys and these mechanisms built into product where they can. You can save like a variation you like. And then we use that as also as part of our feedback loop and, and then you try and make mistakes and fix them. That's basically what we did

Dhaval:

What was the can you unpack the first option that said you have APIs from Google ads and Facebook ads? How would that improve the prompt?

Yaniv Makover:

If you create a copy with Anyword for your Facebook ads and you run those ads, we will look at the performance of that copy, compare it with your other copy on your ads and then see if it did well or not, and then generalize that into some insights about, okay this is not working well. I'll give you an example. So in a Google ad, you can talk about your brands depending on keywords and campaigns. And some brands have really well known really well known. It's it makes total sense for them to actually incorporate way more than that way way more of their brand in their ads and others. It's just all benefits and features of the product and not necessarily the brands, but that you really have to define, let's say, the product itself. So for instance, Anyword, if we had to write an ad for Anyword, we'd have to say AI writer at some point because we're not Nike but for Nike, like I just, the brand itself. So those understandings have to go back into the prompt. And then that's something, you know, from performance and what people like to use.

Dhaval:

So what I'm gleaning from this is for improving the prompts. Feedback loop is critical. Being able to have that feedback from the market, and from the users, helps you improve those prompts. Now that I say it in the hint side, it sounds very obvious, but thank you for helping us glean that. Yeah. Any other advice you have for product creators? Who are either embarking on this journey now that there is, has been, there has been this new technology or anyone who has been doing it for many years and just wanna make that leap. What advice do you have for product creators in this space?

Yaniv Makover:

I think it's super exciting, like exciting times about what's possible at the same time. And I think they're going to be many winners. And big companies are gonna disrupt like existing workflows and industries, I think thinking about specifically what problem you're solving and is it can it what is the incumbent gonna do. So this is a feature in another product. I think that's not easy. I think. I think it I think for different problems, there's different answers. I'd really focus on that. And also I think. Temptation because of the pace of, how the technology is evolving to, to do more and to cover more ground. Oh, let's go into images and let's go into, and you really have to be disciplined because my guess is that if you're not specific enough and narrow enough in your focus, over time, you'll have you won't have the best product. You have kinda like a, and I think it's always a challenge in product management, but I think now it's even more with ai. Specifically, you have to get your user experience right. And and it's very , tempting to just build lots of stuff really fast

Dhaval:

thank you, Yaniv. It was a pleasure having you on the show. Thank you so much for making time for us. Thank you for having me.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard
This former Meta Engineer turned a hackathon product into an AI startup with 150% growth month over month.
8 mars 2023· AI Product Creators
Lilly Chen is the Founder & CEO of Contenda. She is the former Software Engineer at Meta. Contenda's artificial intelligence tools reimagine your content in new formats for your audience to discover, with no extra work from you. In today’s episode, We talk about how they manually labeled their data and created a golden test set for how people viewed content. Lilly also explains how they use AI to transform video into written content publishable on their users' websites without any additional work. She also gives advice to product creators interested in infusing AI into their existing products. If you're interested in building AI-powered content creation or want to learn more about the benefits of using AI in your product. Tune in to hear Lilly's insights and experiences in building Contenda and how you can apply these lessons to your own business.

Where to find Lilly Chen:

• LinkedIn: https://www.linkedin.com/in/lillychen48

Where to find Dhaval:

• Twitter: https://twitter.com/DhavalBhatt

• Instagram: https://www.instagram.com/dhaval.bhatt/

• LinkedIn: https://www.linkedin.com/in/dhavalbhatt

Transcript:-

Dhaval:
This former meta engineer turned a hackathon product into an AI startup with 150% growth month over month. In this episode, we talk about her launch story, how she built the initial team, and her product development journey. Specifically, we discussed an interesting learning lesson around innovative approach she uses for data labeling and creating the golden data set for building your ML model. She shares a few examples of how product creators can get started with building ai. Without deep technical expertise, she also shares a tangible workflow that they can use to do so. Today my guest is Lily Chan. She's a former software engineer at Meta. She's the founder and CEO of Contenda. Contenda is an artificial intelligent tool. Reimagine your workflow in new formats for audience to discover with no extra work from you, and scale your existing technical content faster.

Welcome to the call, Lilly. Thank you for joining us. Tell us about your product.

Lilly:
Contenda Scales, technical content marketing for developer advocates.

Dhaval:
Okay. What does that mean? Tell me. Tell me, that's very interesting. You got that. You nailed it. You nailed the position there.

Lilly:
I mean, the buzzword of the day is generative ai. We do fall in that category. Contenda uses large language models to generate technical content with a high degree of accuracy.

Dhaval:
Wow. Technical content is the hardest content to create of all the content categories, right? It's very easy to spit out sales copy. But technical content that's pretty challenging. So your audience is developer is that right? is a typical developer who wants to create documentation technical documentation. Is that your audience? Tell me a little more about your target user.

Lilly:
Our target users they're called developer advocates, so they write content for other developers. They will oftentimes do a live stream, a lecture, or a conference talk, and then they need to transform that content into a written form, such as a blog tutorial and sometimes documentation.

Dhaval:
Got it. Yeah. So I, as a product person, I have to write a lot of documentation based on. Product launch based on the product demo. And that's the type of a content, you help advocates with. Is that? Did I get there, right? Correct. Okay. Now, tell me where you are in that journey of your product. Is that something you have launched? Are you working on it, et cetera

Lilly:
When we first started building this product, we served everything through email and Google Docs. That meant that you can't sign up on our website to use the product. You just had to email me personally. We spent all of our time building the machine learning and backend infrastructure to do that. Now we have reached the point where our interest form on our website has grown 150% month over month for over the past quarter and a half. And these hundreds of people can't get on our platform unless we build one. So that's what we're currently doing, building a platform.

Dhaval:
Wow. Okay. So you have productized it. To your manual workflows, and right now, you are automating those workflows to create a user experience. How many users do you have? , is that, can someone use your product now? If they want to use it, how many users do you have? How much revenue do you have, et cetera?

Lilly:
We currently only do enterprise-level conversations. That's something that we're looking to change. We would love to get in touch more with developer advocates who have a personal brand and are thinking about using contender for their own Twitch stream or Twitter or blog, and then eventually get to a point where we can say, Hey, would you like to bring our product to work? Would you like to use it on the enterprise level?

Dhaval:
Wow. Okay. So would you say that you have any revenue at this point? Do you have customers that are using this at enterprise level? And, if so, like where does that sweet spotlight for you right now and in the future?

Lilly:
We do have a few enterprise customers. I won't disclose what each person is paying. But the total ACV value is somewhere between 50 k to 100k plus.

Dhaval:
Got it. Is this something that you have bootstrapped? Have you built this on your own? Have you raised a round of funding? Tell us a little bit about how you got to it. Started it.

Lilly:
Well, funny enough, it actually started off as a hackathon project. I was working as a full-time machine learning infrastructure engineer at Meta, and I had just flown out some friends to California for a hackathon project. That hackathon project was for Twitch streaming. It was a retention project for Twitch streamers, and our project went viral. It ended up helping a Twitch streamer break a Guinness World record for most subscribers in a month. It went viral, and a couple of news outlets picked up on the story. So investors reached out from there, and that's how we became a venture-backed business.

Dhaval:
Wow. So what was this hackathon project trying to do? You said something about retention for streamers. Tell me a little more there.

Lilly:
Right So our idea was to build something HubSpot esque, but for content creators on the individual level, Twitch streamers have really, really high churn on their subscribers, and so we wanted to build a product that could help a Twitch streamer retain subscribers over time with the project that we did.

Lilly:
His name was Ludwig. He's. It's currently the Guinness World Record holder for most subscribers on Twitch. We wanted to see if he spent a dollar with us, how much we could retain for him over time. And in the first month, for every dollar he spent with us, he earned a dollar 70 back, and then the following months it would drop off, 50 cents, 20 cents, so on, so forth. But overall, it was very good margins for both of us.

Dhaval:
Wow. So how did you do that? What was the recipe to increase the retention there?

Lilly:
Right. So if you are a Twitch subscriber to Ludwig, you received a notification from us that you could come and fill out this form. On this form, you would be randomly distributed into one of two groups. In one group, you received a digital hello from Ludwig, and in another group, you received a physical sticker in the mail that had a Ludwig emote on it. And we basically ran this ab test to discover that these real-life people, the people that you interacted with through the physical stickers, retained much, much higher over the period of a quarter than the other group did.

Dhaval:
Oh wow. So then this project went viral. You helped that streamer make more money. What happened after that? Like, did you keep the team together, and you continue to build the MVP? Did you raise the around of capital? Tell me a little more. Tell me all the juicy stuff that went into after that.

Lilly:
Yeah, we got our first million-dollar check shortly after that project was released. The funny thing about working with a Guinness World record holder is it only gets worse from there. The market only gets smaller, so pretty quickly, I would say by that summer, we realized this project had no legs in the venture market, and we needed to pivot. The team stayed together. But that's when we ended up discovering. Developer advocates because they were streaming on Twitch under the science and technology section.

Dhaval:
Got it. So you kept the team together. You used that initial funding that you got to pivot towards advocates. And that's what you're, you build a product, you got some more enterprise customers, and now you are trying to further create a UX experience around that so that you can have more, more users. Did I get that product journey right?

Lilly:
Absolutely nailed.

Dhaval:
Awesome. Wonderful. So we got the context of what you were working on. We got the context of what problem you were solving. Let's unpack how you built the initial team. What was the process like? How did you attract those people in your team? And then, before you answer that question, tell us a little bit about your background. Are you a technical co-founder? Are you a technical founder? Are you the, playing the CEO role, the business role? So tell me your role, and then tell me about how you got other people to join this interesting project with you.

Lilly:
I'm the CEO and founder. I have a technical background. I used to be a machine learning infrastructure engineer at Meta as well as a DevOps engineer at Rapid Seven, a public cybersecurity company, and I've been the first software engineering hire at a gaming startup. That being said, I don't have a CS degree. My background in undergrad is actually economics and math. I thought someday I might pursue a Ph.D. in economics. I'm also a high school dropout, so education, formal education-wise, I have very little. That being said, I would say my team is relatively diverse in their backgrounds. Our CTO, Cassidy Williams is a prominent developer advocate. She took over all of the managing of engineers from me, so I am free from managerial work.

Dhaval:
How did you get Ca Cassidy? How did you, how were you able to get her on board? What was that experience like? Did you how did you build a relationship with her to trust you?

Lilly:
I met her husband playing video games. And her husband said you should meet my wife. She's pretty cool. And I was like, yeah, that does sound pretty cool. And that's how I met Cassidy. We started hanging out, just chatting. Funny enough, when I floated her joining the team, I said, wouldn't it be crazy if you like came and worked here? Unless and I sort of made that joke, like I wanna say, over a period of a couple months and at one point she started making the joke back at me and then I thought, wait, are you being serious? And she was trying to figure out if I was being serious. And turns out we were both being serious. And so she joined.

Dhaval:
Wow. Interesting. Very cool. Let's change the gears a little bit there. Tell me a little bit about how you built the current product. Is that built on top of large language models? Is that something that you have built on your own Contenda is a product that helps developer advocates create technical documentation. What was the process of building that product?

Lilly:
So one key thing to note is we only repurpose existing content, meaning we have to work off a source of truth, unlike other LLMs that generate content from the training dataset that they were built off of. We actually generate content from someone's original word. That's really important because we actually have an evaluation model that we run on top of the generative model. So, for example, a generative model creates the original writing. Actually, let me restart on that. Let's take the transcript from a conference talk that you did. We would break that transcript, say, into 10 sections. That first section. We would then put into a prompt as an input, say, transform this into a written paragraph of some sort. That paragraph gets generated 10 times. We score each of those paragraph. With our own in-house evaluation model, that's the determine the quality of the writing, which is subjective. The way that we determine if a number is good or not is we run that same scoring model on all of your existing content. That gives us an idea of what you determine subjectively to be good publishable writing. From there, we take the highest winner, and we feed that and the next section of the transcript to create that next piece and so on and so forth. So that's how we create a blog using both LLMs and our in-house evaluation model.

Dhaval:
Yeah, that was very helpful. One thing I would like to gain further clarity on is how you evaluate. Is there a human in the loop who decides to score in the evaluation model? How? How does that happen?

Lilly:
So what happened when we were delivering our content through email and Google Docs? We asked all of our early testers to leave it as a suggestion in Google Docs. Then we manually went through and labeled all of that data, and that created our golden test set for how people were viewing content. We had a variety of writers from notable different roles, whether it was software development, marketing, VP of engineering, and so on and so forth at multiple companies. This created a data set of below thousands, which we found was sufficient enough to train a classification model from.

Dhaval:
Did you build that classification model on your own, or did you build fine tune an existing LLM for classification?

Lilly:
That is our own model.

Dhaval:
Great. Now, where in this journey, Does AI gets infused? Is it from the very beginning? Tell me a little bit about the user journey. What does that look like? When does a user come in? What is that awareness? When does a user gets delighted with the content? Yeah. If you could paint a picture of user, typical user journey, and where is AI infused in that journey?

Lilly:
Right. When we onboard users, the first thing we do is go collect all of their content data through whatever CMS they're using, and we go ahead and score that using our in-house evaluation model. From there, we have a benchmark of what they consider to be good content. Then they can give us a video of anybody, anybody speaking at all, and we can transform it into writing that they would find publishable on their website without any additional work from them.

Dhaval:
Wow. Okay. That's pretty streamlined workflow. Thank you for sharing all of that. Let's shift the gears a little bit. What advice do you have for creators who are either building AI products from ground up? They're like excited about all this revolution that's happening. They wanna build something. They don't have a whole lot of technical background like you, but they are excited. They have motivation. They wanna build an AI product, or they wanna infuse AI in their existing product. In other words, they work at some company, and they wanna figure out a way to tap into this revolution to be able to improve their user experience. So AI creators, right? How can they start on this journey? What is your advice for them?

Lilly:
I really believe that OpenAI has lowered the barrier of entry so that anybody who is even remotely technical can really attain value from the product. My advice to you would be there's a lot you can do with just the base API of GPT and that you should be looking for low-hanging fruit where you can use the model to do manual tasks that people are performing. Anyway, anything that is highly repetitive yet manual, automating that with an LLM, is probably the lowest hanging fruit to value ratio right now. I think a good example actually would probably be cleaning form data. So let's say you had a survey that you sent out to a bunch of people. People have in the leave more comment section. People will write literally anything in that box. Having a person go through and read all of those could be really hard. I think having an LLM summarize what those things are into a certain category could give you some quantitative data that's really easy for you to work with. So you could just even do something like you could just categorize into sentiment. You could do positive and negative, like figure out if the comment box is that someone saying something, a glowing review like they had more to say that was really positive. Or are they leaving you feedback that they want you to work on because it's negative? Even just seeing that split can give you a better idea of whether or not it's worth spending any time on reading those form submissions.

Dhaval:
Very interesting. So there. Hundreds of opportunities that product creators could be working on. One, you just gave us how to clean up the form data, the survey data. What is your framework of choosing to work on what you choose to work on? How will you approach that? How would you rank the opportunities in this space?

Lilly:
Oh. When we started working on this particular product chat, GPT had not been released yet. Generative AI was not a hype train word, so for us, it purely had to do with our backgrounds. I'm a self-taught developer. I only, I was only able to break into software engineering into machine learning because Really nice people put up free content on the internet. People like who, who write for product hq. You give me your expertise, you tell me your stories, you tell me how to learn things where I can find more. And that ultimately I ended up working at Meta, which is a pretty highly competitive place to find a job for someone with no CS background.

Dhaval:
Interesting. Thank you for sharing all of that. I would love To hear about your thoughts on where is a good starting point or for a product creator who's interested in creating in this space, what is a good starting point for that creator? To learn?

Lilly:
If you are not technical and you want to start exploring some possible business solutions, I would probably recommend no code tools such as Zapier or low code tools such as auto code. In those situations, you can play around with integrations, set it up with a Google sheet, with email, with a CMS that you're working off of, and try finding a problem that you have to solve every single day at your job. Maybe you spend 20 minutes every single day at your job solving this particular problem. Try automating that solution. See if it works for you. If you like it and it solves your problem. In the worst case situation, you just saved yourself 20 minutes every day for the rest of this job. That's fantastic. And in a better case solution, maybe you find out that this other team has everyone has to do that every day for two hours. They have that same exact problem, and then you can find some way to scale it from there.

Dhaval:
Got it. Thank you so much. One thing we'll end on is your vision of your company, your future vision of the product. Where do you think your product is going? Tell us a little bit about the future vision, like five, 10 year vision.

Lilly:
Yeah, we're actually already right now, so we're working on the platform, but we're also working on building extensions with popular CMSs. The idea here is not only does Contenda generate this content for you, but we could dynamically generate it for your audience. We've been talking a lot today about what if I'm technical, what if I'm not technical, and what that content looks like for those different audience groups. So imagine that when you go to look on a blog and you're reading something about a guide, how to break into AI software, building your own company, so and so. Depending on whether or not you're technical, it'll display different resources. That's something that contenda has the ability to do, and we can learn from what the audience wants by dynamically generating content,

Dhaval:
Dynamic content generation for content creators for different audiences or all types of content.

Lilly:
For our customer's customer. So it would be like people like us when we go and view content on a certain landing page. It could dynamically adjust the resources and fill in the blanks for what makes the most sense for our background uniquely.

Dhaval:
Wow. Okay. Highly personalized, individualized content for the reader or the consumer impressive. I wish you all the best in that journey. Thank you so much for making time for this show. I'm looking forward to hear back from you in a future podcast where you know you come back and share more learning lessons with us. Thank you, Lilly.

Lilly:
Thanks for having me.
- Écouter Écoute encore Continuer Écoutez...
- Écoutez plus tard Écoutez plus tard

Episodes

Transforming Creativity and Content Creation with Zach Hanson - A Deep Dive into the Future of AI Product Management

This founder built a platform to create lifelike characters for immersive experiences | EXCLUSIVE!

How this founder built a personal language model to mimic your behavior, knowledge, and style.

What happens when a PhD Professor in Analytics launches an AI Writing Product?

This former Dropbox engineer built an AI Product to create stylized eCommerce product photos

How founder this build a website copy and image generation plugin

This founder built an AI product to replace your employer with AI.

This founder built an AI Writing Product that serves 4 million customers and got acquired in 2 years.

This founder built an AI game that attracted 100K users in its first week of launch.

This founder built an AI product to help writers beat creative block and generates over $750k ARR.

How they built an AI Product where you can discover and generate art to 100x your creativity.

This founder built an AI product to help you become a super-learner.

This award-winning founder built an AI Design Product and a user community of half a million users in less than 5 years

This founder built an AI Content Workflow product and grew it from 0 to 1M users in 1 Year.

This Founder Harnessed AI Language Models in Just 2 Years, Revolutionizing Copywriting with 850,000 Users and Exponential Growth

This former Meta Engineer turned a hackathon product into an AI startup with 150% growth month over month.