You you you you you you you you you you you you You You Hey.

I could see that some people have been having trouble hearing me.

If you can hear my voice, give me a thumbs up.

Let me know that the audio is coming through for you.

Blaine, my man, glad that you're here and can hear me.

Can somebody just go into our private chat for speakers and let Ido know, yes, we can hear, everything's coming in.

And if you can't hear us at some point in the session, Twitter has been super buggy as you know lately, and it just means that what we need to do is either restart your system or sometimes you just need to pause.

If you see that I cannot hear something it is because there's probably a Twitter issue that's affecting me and I've got Salil here he is our co-host he will be just jumping in and taking over.

I'm gonna invite you Celil to be a speaker And co-host a session with us fantastic.

Thank you all for being here If you can hit that bottom right chat button and just retweet this session let people know that we're live letting them know we're live brings in more people into the conversation and allows them to frankly sometimes add some insight privately to us via direct messages, sometimes connect with us, and sometimes just bring in their followers which allows us to broaden out this conversation.

Some of you know me as the person who's been doing interviews on Mixergy for years, over 2200 of them on the site.

And I regret that I did not do enough interviews about AI because man, I've been following Bhanu.

Bhanu, do I am I pronouncing your name right?

Bhanu created SiteGPT and I've just been watching him over the years be an indie developer, cranking things out by himself.

And because he's doing this all in public, we're all getting to watch it.

And I think he gets some company and gets some customers from it.

And I've been watching this guy take this idea for SiteGPT from 0 to just an incredible number of users, incredible revenue over a short period of time.

I think February is when he first started talking about it.

And then I talked to Blaine because I love this tool called Cast Magic, which basically takes my whole podcast, transcribes it better than any other system I've ever used before, and then gives me 10 suggestions for headlines, gives me suggestions for tweets.

I'll be honest with all of you, it's not magical.

It doesn't end up being like the perfect writing, but sometimes I feel like, hey, that's good.

And sometimes I feel like that's terrible.

What he should have said is, and then I come

I could write my own prompts for it and then come up with my own ways of extracting information from my podcast.

I won't reveal it, he told me the numbers behind it, how many people are using it for not just podcasting, but for meeting notes to be able to extract information for meetings.

Brian has been running Heights Platform for years, enabling creators to create content that people actually pay for and participate with.

And he's messing around with it and adding functionality, not to a brand new tool that's just allowed because of AI, but adding functionality to a tool that's worked well, that now has even bigger super power because of this.

I wanna know about what I've been missing.

What is it that's going on in this space?

And so I put together this Twitter space to understand what you all are doing, especially the indie side.

I get pitched so many times by entrepreneurs who wanna do, who have raised 100 plus million dollars and now they're making a million dollars a month.

I wanna know about the indie guys, the indie people who are building companies.

That's what we're gonna be finding out about.

And I thought for the first introductions, what we could do is just kind of go around and have each 1 of you who's a speaker, first of all, ask for speaking.

And second, just unmute yourselves and in a sentence or 2, don't go too long.

Tell us what what your tool is and what it does.

And then we'll keep coming up with more questions and ways of engaging everyone.

But Bonnie, since I brought you up, if you feel comfortable unmuting, tell us what your tool is and what it does and then everyone else just take from there.

Yeah, so what I built it's called SiteGPT.

It's a way to create a chart box based on your website content right like all the all my customers are use it mainly for customer support.

This replaces, this will act as a first step for providing customer support.

You just give it your website link and it automatically gets trained on all the links in your website.

Do you mind giving us the revenue numbers?

And if anyone wants to give us a sense of how many users you have or anything about growth, that'd be helpful too.

Kevin, we'll let you go next, but Bhanu, do you mind just giving the revenue also?

So I'm Kevin, the founder of the Creator Kit.

We do AI generated photos and videos for e-commerce marketing.

And we specialize in doing basically media for marketing and e-commerce specifically.

Happy to chat more then I'll let you everyone introduce themselves.

We are working on ChatterDocs, which is an advanced custom chatbot builder built on OpenAI like a lot of these other folks.

We allow you to ingest sources from not only like your website, but things like Notion, YouTube, MP3, RSS, all that kind of stuff.

And so lots of different ways to get your data in.

And then on the other side, there's a lot of different ways to get it out via the API, Slack, embedding it on your website.

And we also have like custom landing pages that you could create as well.

I think the biggest difference that we bring to

the table right now is really just this, the ability to customize the workflow.

So you get the ability to kind of say, pick a direction when somebody's clicking around or interacting with the bot and be able to schedule appointments and collect lead data and stuff like that.

How about Brian next and then Blaine after that?

Hey guys, I'm Brian McAnulty, the founder of Heights Platform.

We're an online course creation software and community building software.

And When it comes to AI, we added some features called Heights AI, which allows our creators to be able to ask things like support questions about how to use the software, be able to ask questions about how to grow their business.

Which is all based on like our, our knowledge, our content that our teams created over the years.

And more recently, the Heights AI can act as kind of like a co-pilot feature where you can actually ask it to make changes to your product, say like publish my guitar course, make the launch date 30 days from now, and then set the price to

So what I think is really exciting is now people don't even have to learn how to use the interface.

I'm 1 of the founders of Cast Magic and we are an AI voice to content platform.

So basically we allow for customized generative content and workflows for any sort of conversational audio.

We started the platform as a podcast AI copywriting tool, basically before seeing a more horizontal opportunity for any sort of audio.

So kind of as Andrew alluded to, you can use it for anything from podcasts to meetings, to voice memos, to pretty much anything.

It allows you to chat with any document of any type in any language within seconds.

I mean, the biggest features we have is we immediately generate summaries, FAQs, which are very important for people, highlights of documents and we are coming out with an API soon so that you can embed all this magic within your own tool or website.

Hey everyone, Sebastian from Hey Milo here.

Me and my co-founder, we built generative AI-powered voice agents to handle technical interviews for companies and vet candidates at scale more effectively.

Our AI agents can actually call candidates, do an interview with them, respond to their input, and then at the end, provide a summary on their fit for the role.

So we're looking to promote more inclusive hiring, more effective hiring.

And we launched, I'd say, a week ago now.

And so we're talking to a bunch of customers and looking to scale to different companies.

Yeah, it sounds beautiful in the demo on your website.

It's like a person, We know that it's not a real person, but it sounds like a person asking questions based on people's resume, asking follow-up questions.

It's a really cool demo and a cool product.

I've been in the startup space for several years.

Met Andrew a couple of years ago in the blockchain world.

I'm actively involved with a few different projects in the AI startup space, building a tool which I like to call Postman for prompts.

So essentially how you create and refine and test APIs.

You can create, refine, test, share, collaborate on prompts.

So building that, it's called Zen Prompt and also working on an AI newsletter for founders and funders.

So those are my 2 projects, but been in the startup space both as an entrepreneur and as an advisor for over 5 years at this point.

So I thought we could start with 1 thing that's really easy in working with AI, especially with OpenAI's API access.

And then maybe we could go to something that's a little bit challenging in the space and be open about the challenges and if there's a way that you've overcome it, share that.

But Atul, I know we're gonna oversimplify it, but Repl.it did a quick story about how you built your app.

Do you feel comfortable sharing how much it cost you and give us the short version of

it and then we can go into the more complicated stuff, but give us a short version of what worked well and easily.

So I had this clear vision of what I wanted to build and I also had the architecture laid out and I was I was knee deep in my own work at at Amazon and so I didn't have the time to go learn the coding and like do it.

I went through a couple of developers who are not actually able to take the vision and and complete it in in the way I wanted.

But eventually I hit it with the developer on the replit and he was exactly able to build what I wanted and it cost me around $3, 000 in total.

And then He and me have now collaborated and we are punching out features every 2 weeks.

And do you feel comfortable saying what it cost to get the whole thing built so far?

Bhanu, is there something that worked out fairly easily because of the technology that you have access to today?

Yeah, so I think I knew nothing about AI before this, but OpenAI made it so accessible to use AI like it's just a bunch of APIs now like anyone who is a good software developer can build AI applications on top of all these APIs.

So yeah, so it made many things possible OpenAI.

You are giving people the ability to add chat GPT that focuses on content other on their site can you give us just a quick overview of how you're able to do it using openai

so yeah openai has something called impedance so what it does is it gives you similarity between uh 2 pieces of text right like if you give it 2 pieces it will be able to tell whether these 2 pieces are similar or these 2 other pieces are similar it will be able to compare them so that that's what open AI does So I use that that tech to build this.

I scrape the entire website, I get all the content from the website, and then I build embeddings out of it.

And I use those embeddings to answer the user question.

Okay, and I did share for both of the 2 recent speakers some tweets that show how they did it and how they're doing.

Lane, it feels to me like Cast Magic is incredibly effective, but maybe I'm oversimplifying my understanding of the back end.

But it seems like what you're doing is you're having a transcribed and there are multiple sources where you could have a transcribed.

And then you're putting it into chat GPT and having your own prompts pull in data for people.

Yeah, I think that's a pretty accurate way to describe it.

I think what goes on in the background though, is there's so many corner cases and use cases that you have to solve for, for all the different types of audio, right?

So, you know, there's like a Twitter space where there's 20 different speakers is a different scenario than a dialogue versus a monologue.

And 1 of the challenges when working with longer format sort of content is the context windows that you're allowed with OpenAI.

But that was kind of initially the reason we built the service was because you couldn't just take a, you know, 20, 30, 40 page transcript, drop it into chat GPT and start asking it questions.

So there's a bunch of smart processing and stuff like that that goes on the background, being able to handle a bunch of different APIs for all the different languages and everything.

So that's kind of where the software and workflows that we build come in.

Because I'm a podcaster who has his interviews transcribed regularly.

I tried putting it into just a chat GPT window.

I then started breaking it up into 4 different clips and I would say, I'm about to send you 4 different clips, just study these and then I'll ask you questions afterwards.

What did you do that I wasn't able to do by just using the chat interface to be able to query long form content?

Yeah, so the chat versus the API are 2 different like products so what you're able to do in chat versus what you're able to do in the API are a little bit different and then like I said before we're able to you know, basically take basically summarize and process and take all the information that's in the transcript, store it in different ways in different places, and then be able to, again, query against that to generate accurate results.

So that's why when you're asking Cast Magic for specific content related to a really long thing, it's able to understand what you're looking for, go find it, identify it, and then generate content based on that.

And Salil, as my co-host here, you have a bigger, deeper understanding of the tech here, so feel free to jump in at any time.

Just unmute yourself and ask follow-up questions.

Brian, was there anything that worked fairly easily?

Was there like an easy win for you in adding AI?

And then if someone else wants to go, just give me the thumbs up so that you can go next.

Sure, I think what was surprising to me at least was the ability for us to get just all of our content in there of our support content, about how to use our software, our marketing content, and being able to have that as a tool that customers could then ask, where it's not gonna be 100% as good as actually asking somebody on our team, but the fact that it gives you information instantly means that if it's even 80 percent, that's going to be so useful to people.

I guess coming from that and working in that, what was so interesting is finding that like with all of these, the AI tools now you're writing these prompts in English.

And so my, my team's all helping write prompts.

I have everybody helping and working on that, even if they're not a technical role, partially because I want them to be able to understand it, but also because it's really possible.

If you know how to write instructions out, then you know how to kind of build with AI, even if you're not a developer.

And then Kevin, I saw you come in next after that.

Yeah, so I, like Bahu, I don't have like a ton of experience working in AI, of course.

I think coming into this was just about learning how to stitch together the APIs.

But I thought what was really interesting is just the tooling around it has developed so quickly.

Like, LangChain is probably 1 of those projects that, I mean, everybody in here is probably a little bit familiar with, but it's so powerful because it just gives you the ability to swap out different tools easily and test lots of things.

Like you can just test different prompts, you can test different backend providers on the embedding side, and you can test out different ways to interact with the content itself whether you do it with the chat style or if you use the API for more of the summarization kind of things like Blaine was talking about.

So I think that's what's been really fun and exciting for me is just to see the progression of all the tools around the picks and shovels or whatever, around building.

Because This gets to 1 of the difficulties too in this space, as things move so fast that it's really difficult to keep up.

And so you kind of have to always just be paying attention to what's out there to a certain degree, even if you don't plan on implementing it right away, just because there's probably a better way to do things than what you're doing today, like the next week.

Where are you finding the tools and the resources that let you stay up to date and find out what's going to be amazing tomorrow or is worth looking at?

I mean, I pay attention to a lot of newsletters and unfortunately, this is probably another idea is just to grab all the AI newsletters and look for duplicate content and really just digest it into 1.

But I follow a lot of AI newsletters, Ben's Bites, TLDR, there's a bunch of them that I follow right now.

And just tracking those down, taking a look at them, trying to be good at digesting the information quickly and seeing like, is this gonna be useful for the project?

And then usually just a little bit of a back and forth with my co-founder Ben, who we're chatting with, or that I work with, and we'll just chat about it a little bit and see if it's something that we might be able to use.

And sometimes it's in the form of code snippets that you can find in the open source repos and other times it's just an article that helps you think about a different way to attack a certain problem.

So yeah, it's just staying close to those things.

And then also paying attention on Twitter, when people are tweeting, like following all the main folks there and staying up to date with notifications from those people.

Kevin, Jim button in here, I saw that you're next.

And then, hey, Milo, if you want to come in after that, that'd be a good time.

So for us, our journey with AI started quite a few years ago before, before it was, before the AI boom, we were integrating many of the AI models on, for example, for background removal and that kind of like features that we had on our AI video tour that we launched 4 years ago.

So we had a lot of features that weigh AI power.

So it wasn't for us kind of like when the AI boom started, we weren't piecing together many APIs, which I believe there's a huge opportunity to piece together the right APIs with the right embeddings.

So for those who are not technical, we can get into what the embeddings are.

But For us, we just went very deep and technical with my co-founder into actually creating our own model of stable diffusion so that Stable diffusion is the model to create AI images and we created our own version of StableDiffusion or forking StableDiffusion so that we can optimize it for the specific use case of generating product images.

So when you try StableDiffusion or any of the AI tools out there for product images for specific uses of, a specific use case of creating product images, what you'll get is that usually it will deform the product or for that, that the, that brands are trying to use for, for marketing.

So that's why we decided to focus on fixing that specific use case and going deep into what's called the diffusion models.

And so we went deep technical with my co-founder instead of the path of piecing together the right APIs.

And it turned out, it turned out a really good outcome in the sense that we are kind of the only ones that have the this possibility of creating product images that are not deforming the product.

Kevin, in the beginning, when you had Creator Kit launched, what did you do and what powered that?

So when we started, we started as a video creator, purpose-built for e-commerce, and we started 4 years ago, specifically building a video editing tool for e-commerce.

When we saw that most of the e-commerce stores required product images for their videos, then that's when we saw that we wanted to integrate an API.

Just like John said, we wanted to piece together the right APIs when Stable Diffusion launched and it was the boom of the AI-generated images, we just said, oh, you can just integrate this API into our platform and it will work really well to generate product images.

But it turned out that there wasn't any API available because most of the available tools were deforming the product images.

So that's when we went deep into creating something that's why it was very technical and required quite a lot of work.

I don't know if I would recommend the path of going very, very technical instead of just piecing together a few of the available APIs into a very, very specific workflow that you know that customers will make sense.

Kevin, with Mid-Journey making all these updates so rapidly, right, so I think 5.2 just came out a few days ago, there was 4 or 5, it's basically in the last 6 months, the whole mid journey model has improved significantly.

Do you think at some point, I know they don't have an API right now, at some point you might want to just hook into mid journey because that's like a foundation model, right?

I feel a lot of these products will be built on these industrial grade foundation models, or do you think the approach that you're following, which is bring home the data, bring home the model, build it in-house, Any thoughts on future approaches, assuming a stable diffusion like foundation model becomes publicly available via API?

Sure, so the thing is that for us, we are tackling a very, very specific use case that Midjourney is not tackling, which is product images with AI.

And you cannot do product image with AI in Mid Journey because it will deform the product image because the AI model is not designed so that it's not deforming the edges.

And you can use the original product image with your product and embed it into an image that is embedded in a way that it doesn't look generated.

That's something that Midjourney is creating right now, the best AI generated images.

But it doesn't work for the use case of uploading your product image with white background and then generating product images.

So this is a very, very specific use case that again, it's a very huge market, but a very specific use case that make journeys is not useful for it right now.

So the way we see it is that you need to make the right changes.

And if you want to tackle a specific use case in the diffusion model area, There's so many areas that there are many, for example, for the ones who are working on AI audio, I believe that the huge opportunity is actually on the data that they are unlocking that is not available for the open AI, for example.

Yeah, I'll actually like maybe chime in there.

Like I completely agree with like getting like your own kind of custom data to fine tune the model.

But I would take actually, for our use case at least, the opposite approach where generative AI has allowed us to have that foundational model where we no longer have to, from scratch, train like a custom model to do certain things.

And so that's kind of allowed us to like iterate on a number of different generative AI or AI products over the last 4 months, which previously we honestly couldn't do for like 2 key reasons.

Either it would cost too much or it would take too long.

And so I think that's been like a really, that's been like the thing that's allowed a lot of, I think, entrepreneurs and founders and developers to move so quickly.

It's like the fact that you don't have to now collect a ton of training data, build a model, spend money on the infra for all that.

And then in terms of just our journey with HeyMilo, we actually pivoted from another kind of startup or product just 3 weeks ago, and we built this in just a couple of weeks, which really could not have been done before.

And we've been able to iterate and go to market really, really quickly.

And I think that's something that really stood out to us.

And in terms of tech stack, we're using GVT under the hood for the model and then Microsoft Azure for the text to speech.

And then yeah, we're collecting custom data to continue to fine tune the model over time and build a mode like that.

So on the mode piece, and I think that's a very interesting thing, right?

lot of us, right, like indeed developers, we are able to build things like you said, because there are foundation models, right?

ChatGPT is around, they spend multiple years, millions of dollars training the model and now we can spin it up.

Would the mode actually be technical or would the mode be more business driven, right?

It's a social network, so you know, make these multiplayer games or multiplayer apps rather than single player apps.

So those are the modes and not technical modes, per se.

Like I think you can have both the technical mode as well as the business mode.

Like I think custom tuning a model is still going to be better than the generic model.

So you're going to outperform 1 of these generic models still.

And then applying that and integrating into various other use cases is I guess where the business mode comes in.

And that's how I think at HeyMilo, that's how we're thinking about that.

Like, how do we integrate seamlessly into the recruiting stack?

Which is something that like a chat GBT, 1 of these single applications won't be able to do.

See all Unknown transcripts on Generic

12c88615950d7dc3ad95e70705a201a7