Using OSS models to build AI apps with millions of users

Talk transcript

1. 0:00

   [upbeat music] Hello, everybody.

2. 0:16

   Thank you for, uh, coming. My name is Hassan. I'm super excited to be here today, uh, to talk to you about, uh, how to use open source models to build AI apps with millions of users.

3. 0:27

   Um, I've been building AI apps for, for several years now, and I just wanna come tell you a little bit about my journey, show off some of the apps, and then tell you about, uh, kinda my whole process, uh, of, of building these apps from idea to having the finished app and, uh, how I, uh, market them

4. 0:43

   as well, and then some, some, uh, a-advice overall. So I'll start off with some intros about who I am, Together AI, the company I work for, uh, why it's one of the best times to build right now, and then we'll get into demos.

5. 0:54

   Uh, I'll talk about my tech stack and architecture, and then I'll end with, uh, my process for building these apps and, uh, advice I have for other people, uh, building these apps.

6. 1:04

   So, um, I lead developer relations for an AI infra company called, uh, Together AI. But probably the most relevant part, uh, for this talk is the fact that I, I build a lot of these AI apps, and I've probably averaged one a month for the last four years.

7. 1:18

   Um, and I've been fortunate enough that some of them have done, uh, pretty well and, and have gotten, uh, a good amount of users and, and GitHub stars as well.

8. 1:25

   Everything I do also is open source, so I'm gonna show off some demos. Uh, but also you can check those out in your own time, and you can, uh, dig into the code, uh, as well.

9. 1:35

   Um, like I said, I, I lead developer relations for an AI cloud platform called Together AI. Um, we're a platform for open source models. We give you an inference API to be able to query, uh, really any open source models probably that are out there that, that are good.

10. 1:49

    Uh, so we have, uh, chat models like Qwen 3. We have, uh, reasoning models like DeepSeek-R1. We have image models like the Flux context model that came out recently.

11. 1:57

    Uh, we have vision models, audio models, uh, you know, uh, almost all modalities. We also let you, uh, run these models on dedicated instances, uh, or fine-tune them on your own data.

12. 2:07

    And, and we have a GPU cluster product as well, uh, if you want a training cluster.

13. 2:12

    Um, so before I get into the demos, I just wanna talk about the fact that it's, it's really a historic time right now for building, and I think this is because of two things.

14. 2:21

    Like one, I think the barrier for building has lowered dramatically with a lot of these AI tools, right? You can use Cursor and Windsurf with your IDE to autocomplete stuff.

15. 2:31

    You could use, uh, these AI builders like Bolt and v0 and, and Lovable to prototype apps really quickly. Uh, you can use chat apps to kind of brainstorm ideas or, or, um, go through the process of building an app.

16. 2:43

    You know, you can go to a chat app and say, uh, "I'm building a Swift app for the first time. Can you guide me through how to do it?"

17. 2:48

    Right? So I think, like, the, the barrier has lowered on one side. And on the other side, we're having these, like, groundbreaking models come out every single week that are enabling, um, completely-- en-enabling builders to, to build completely new things.

18. 3:01

    So it's like these incredible AI models are coming out that you can build with, and it's gotten a little bit easier. Um, so I, I, I just think it's, it's an incredible time to, to be building apps right now.

19. 3:12

    Um, I'm going to dig into a couple of demos. These are some of the apps I, I've built that I'll, I'll spend like thirty seconds [chuckles] on each one demo-demoing really quickly.

20. 3:20

    Uh, but one thing to, to, to note really quick is the fact that some of these apps have a couple thousand, uh, people that have used them, and then some of them [chuckles] have like a million plus.

21. 3:27

    And I think, like, uh, at, at a certain point it is a numbers game and you kinda have to try out a lot of different stuff, and, and I'll talk about kinda how, how I approach that.

22. 3:35

    Um, so we'll, uh, switch into some demos. So in the beginning when I was-- when I started building kind of a few years ago, um, I wouldn't know what to build.

23. 3:43

    So I'd go on Twitter, and I'd look up, you know, people that tweeted like, "Oh, I need an app to do this." And so here Samina said, uh, "Can someone build an AI app to help me pick some glasses?"

24. 3:51

    And I was like, "Sure, I got you," um, and, uh, you know, built this little, um, tool that takes a bunch of different requirements, uh, turns them into a query, and then actually, um, uses the Amazon API to find products that you can actually, you know, click on and, and buy.

25. 4:09

    Um, an-another one of these was, uh, one of my friends, Theo, tweeted out like, uh, you know, "I just want an AI app that writes my commit messages for me."

26. 4:16

    And, uh, my CTO at the time was like, "I love that idea." And, uh, so I just decided to build it. And, um, all it does is when you do Git add, it takes the Git diff, it sends it to an a- to an AI model, and, uh, it writes a commit message for you, and it, and

27. 4:27

    it kind of, uh, shows it to you, and you can kind of accept it. So this was one of my first apps that I built that, that ended up doing really well and, uh, got, uh, I, I think it got about forty thousand people that have installed it and, and downloaded it, and it's also open source, and,

28. 4:40

    and I had, uh, a bunch of people contribute, which, which has, uh, been great. Um, another app I built was a, like, text-to-app builder. There's a bunch of these, um, where you can type in an app that you wanna build.

29. 4:52

    Here I'm gonna say, like, a quiz app about American history. And the way this works is it, it takes this prompt, it sends it to an AI model to come up with a project plan, and then it sends it to another model to actually, uh, write React code that, that we can show.

30. 5:04

    So, um, in, in a second this should populate, but, uh, uh, obviously also, also open source. Um, I've had about five million requests go through this app and about a million apps built, um, just through this.

31. 5:16

    Uh, a-and I have, uh, a little over a million people that have ended up, uh, using it. So this is an app, yeah, that, that kinda just generates images, uh, as, as you go through it and, um, also has about o-over a million people and, and has, uh, about forty-eight million images that were actually generated, uh, w-with

32. 5:32

    this app. And I have an app called Napkins where, uh, it, it takes a, uh, screenshot of an, uh... The idea is it takes a screenshot of a napkin.

33. 5:40

    The idea is, like, you draw a little web app on a napkin, it takes a screenshot of it, and it can actually, uh, uh, build it for you. Um, so yeah, that's, uh, that's another one of my apps.

34. 5:50

    That one has about forty thousand, uh, uh, users. I have another one that lets people kind of upload a resume and then builds them, like, a personal site like this, uh, for, for people to check out.

35. 6:00

    Uh, and, you know, there's, there's a bunch of others. Ones that visualize menus. Um, I built like an, an AI, uh, chat app, like a, a tutor app where you can, um, you know, put in a topic that you're interested in, like personal finance, and it can explain it in the level of like an elementary school, um,

36. 6:15

    uh, uh, person. Uh, and then some OCR stuff as well. So I'll get back to the slides. Um,

37. 6:24

    a lot of these apps have, uh, a very similar architecture that I end up using. And so this similar architecture starts with, uh, a user coming in and typing something or uploading an image usually.

38. 6:36

    So it's some sort of user input, right? And I take that, and I generally will send it to an AI model on, on Together AI. So it could be like an image-to-image app where I, I take an image, and I send it to an AI model to make another image and then show that to the user.

39. 6:49

    Um, so step two is I send it to an AI model. Step three, a lot of the time, is I, I store that, uh, image or, or that text in the database to be able to show it to the user.

40. 6:57

    And then step four is I just show it to the user. Um, so it's a, a very, very simple architecture. A lot of the time, there's one single API call happening.

41. 7:05

    You know? The user does something, I send one API call to an AI model, I get the response back, and I show it to the user. Um, and I think it's this simplicity that, uh, is, is really, really important both for moving very quickly, um, but also for validating these ideas.

42. 7:20

    You know, the more simple you can drill down an idea, the, the, the faster you can kinda build. And then this is my tech stack, uh, that I usually use.

43. 7:26

    I use Together AI for all of my AI models. I use Next.js and TypeScript as my, like, full stack framework, uh, for building things. I use Neon as a, uh, as my database.

44. 7:36

    It's, uh, just like a really good serverless Postgres host. I use Clerk for authentication. Uh, I use Prisma as a way to talk to my database in TypeScript. Uh, I use Shadcn and Tailwind for, for styling, S3 for uploading images, Plausible for, uh, website analytics, and I kind of, uh, showed you guys this, uh, in, in a

45. 7:54

    couple different places. Uh, so it's cool that I can see the number of unique visitors. I can also see where they're coming from and what countries and what device they're using it on.

46. 8:02

    Uh, and one thing I always get very, um, uh, surprised by is the fact that a lot of people use most of my apps on, on mobile. So, like, it just goes to show that the, like, mobile experience is really, really important.

47. 8:12

    Um, so Plausible for website analytics, Helicone for LLM analytics, so I can dig into my LLM requests and, like, troubleshoot things, and then Vercel for, uh, for hosting, uh, these apps.

48. 8:24

    Cool. So I'm going to talk about my process for building these apps. And, and a question I get a lot is like, "How do you come up with these ideas?"

49. 8:32

    Uh, and, and I think, like, the, the biggest thing I did for ideas that, that helped me out was just to keep a l- a list of running ideas.

50. 8:41

    Um, I think we all get great ideas at random times, and most of us don't write them down, and I think, like, uh, that makes a lot of the difference.

51. 8:49

    And so I think keeping this list of ideas, uh, trying to write down any time you see an interesting idea or see an interesting product and you say, "Oh, wow, that's actually really cool.

52. 8:56

    Maybe I can use, uh, a similar methodology to... a- and apply it to something else." Um, so ideation is like, uh, a, a big thing and, uh, the list of ideas I have, I usually always have this short list of like the top five.

53. 9:08

    And so I-- like, I know right now kinda the top five apps I wanna build next. Um, and then if anything kind of drops in between, then I'll, I'll build it.

54. 9:17

    You know, if a new image model comes out next week that's like really, really incredible and open source, um, I'm probably gonna be scrambling to build an app with, with that.

55. 9:24

    Uh, naming is another big one. You know, you want a short, memorable name. I-- y- you know, you can use AI tools like DomainsGPT to kind of, uh, check names that also have the domain name available, uh, to, to use them.

56. 9:35

    Uh, number three is design. You know, thinking through how the app will work. Like, you know, I'm gonna have a landing page, and the user's gonna click, like, Enter, and then enter-- get this page where they upload something, and then they see an image.

57. 9:44

    You know? And so it's like these two screens, and, uh, you can either kinda sketch it out on a piece of paper, you can use Figma, uh, or you can use a lot of these prototyping tools, uh, to, to try to help you think through, uh, what, what the app is.

58. 9:56

    Uh, and then I go about building the actual app, so trying to make the simplest possible working version of a specific app. Um, and like I said, I, I always try to shoot for like one API endpoint.

59. 10:08

    You know, like one-- like just very, very simple. Um, step five is like, you know, I have a working prototype now. I, I start to think through, uh, authentication and limits of like, okay, like how expensive, uh, is this?

60. 10:19

    Like, how many, um, how many uses, uses do I wanna give each person per day? Do I wanna, wanna add authentication? Do I wanna add bring your own API key?

61. 10:28

    Uh, so it really just depends on, on the app and the AI model that I'm using. Uh, and then step six is usually kinda prepping for the launch, getting a nice OG image, getting a domain, adding analytics, uh, writing a nice README 'cause like I said, uh, uh, everything I kinda do is open source, so I also

62. 10:41

    wanna make sure the code is, is really easy to use and, and really easy to, uh, to clone. Um, and then the last step is actually launching. And so I usually use LinkedIn or X to, to launch and, uh, you know, kind of, uh, just see what people say a-at that point.

63. 10:59

    Cool. Final section, uh, and then I think we'll get into some Q&A. We have a little bit more time 'cause the demos didn't work. Um, so advice for building apps.

64. 11:07

    I have these like seven tips that I'm gonna go through. Uh, one is thinking of an idea that excites you but is very, very simple, uh, that you should be able to describe to anyone in five words, right?

65. 11:17

    Like BlinkShot, generate real-time images. LlamaCoder, go from some text to an app, right? Like, uh, and I think this is one of the biggest mistakes people use is they try to, they try to think like, "Okay, you know what?

66. 11:27

    I like, I wanna build this, like, personal CRM software that has this dashboard that will email me every week and do this and that," and, you know, they come up with this like grandiose version and spend like six months building it, and then they realize, "Oh, wow, it's like, it's, it's, um, this isn't, this wasn't the right

67. 11:41

    thing to build," or, um, "This is just like really hard and nobody cares about it now." And so it, it's all about, uh, thinking of an idea that excites you but is also, uh, really, really simple.

68. 11:51

    Number two, and this is very underrated, is making sure the UI looks good. A lot of the apps I showed you are AI apps, but I actually spend 80% of the time on the UI for most of my apps.

69. 12:02

    Um, and that should tell you, like, how, how important this thing is. I, I've like... The first few apps I built kind of looked really, really bad, and that was a big part of why nobody kinda used them.

70. 12:11

    And I've started learning that even if you take a really simple idea or something that's like so simple, like summarizing a PDF, right? You can go to ChatGPT and summarize a PDF right now, but I, I built a PDF summary app, and I spent a lot of time making it look really, really, really good, and, and then

71. 12:25

    I had tens of thousands of people that, that ended up using it. Um, so it's all about making it look really good, making it really easy to use, uh, and making it really, really straightforward.

72. 12:34

    Um, the third tip is just keeping the app simple. I talked about this a bunch, but, uh, most of my apps have only one or two API calls. Um, tip number four is trying to incorporate the latest AI models.

73. 12:45

    A lot of these apps, um, that, that I showed you have, like, used some of the latest AI models. Like BlinkShot real-time image generation app, um, that used, uh, a model called Flux Schnell, and I, and I launched this app, I think, two days after this model came out.

74. 13:00

    Um, and so it was one of the first apps that leveraged kinda this new technology or this new really, really fast model that was good. Um, and so doing that kind of increases, uh, the potential for, uh, virality.

75. 13:12

    Uh, tip five is launching early and then iterating. Um, for a lot of these apps, like still to this day, you know, like 40 apps later, I still have no idea what will do really, really well versus what won't.

76. 13:22

    Um, and the only way to de-risk this is to simplify your project, to launch early, and then if nobody really cares about it or not a lot of people use it, then at least, you know, you didn't spend six months building, you spent a week building, and you can kinda move on to another idea.

77. 13:36

    Cool. Uh, another tip that I, that I've kind of done is trying to make it free and open source, um, so folks can kinda learn from you and are, are also very incentivized to share it.

78. 13:45

    Um, and so, uh, th- th- this is something that, that's worked well, uh, for me.

79. 13:50

    And then the last tip I have is, uh, just keep shipping. Um, a lot of AI apps, like I said, don't do very well. It's a numbers game. You have to kind of keep building and building and building, and the more you build, the more you realize, you know, kinda what resonates and what doesn't.

80. 14:03

    The faster you get at building, uh, and the better you get at, like, picking, picking ideas. So, um, a, a lot of it just comes down to, you know, putting in the hours, building a lot of stuff, and then, uh, seeing what happens.

81. 14:16

    And that's all I have. We actually have five minutes for Q&A if anybody has any questions. Um, and yeah, I think you can line up at these, uh, podiums, and we can, uh, take a few questions, but just, uh, before that, uh, you can find me on Twitter [REDACTED:username] or, uh, my email [REDACTED:email_address].

82. 14:34

    Uh, we also have a Together AI booth here at S25. Um, I'll be there, uh, for a few hours after, after the talk, so, uh, come and talk to us.

83. 14:42

    Yes.

84. 14:42

    Okay. Thanks for sharing. I'll pull this on. Uh, yeah, I'll just yell.

85. 14:48

    Great.

86. 14:48

    Uh, oh, it's on. Okay. Uh, thanks for sharing. This is great. Uh, I'm just curious. Uh, you have a lot of traction success, more than many most YC startups.

87. 14:56

    Why don't, why don't you just start a company? It, it seems like the perfect formula. I'm just really curious.

88. 15:00

    Yeah. Um, that's a good question. I mean, a part of it is I, I really enjoy, um, I really enjoy teaching, and a lot of these apps, like I can...

89. 15:08

    A, a lot of these apps, first of all, do well because they're free and open source, right? And so because I'm, I'm strictly not trying to monetize them, I'm, I'm strictly, like, trying to make them free, and I get companies to kinda sponsor it and, and, uh, and, and keep it open source.

90. 15:21

    Um, so I just kind of enjoy doing that and launching these apps, and I, I will also, like, make videos and blog posts about different things. Uh, but yeah, maybe eventually I'll, I'll come out with an idea that I really, really love working on and, uh, and, and do that.

91. 15:34

    But, um, for now, it's just been really fun getting to build a lot of the stuff and experiment.

92. 15:37

    Well, if you ever do start a startup, let me know. I'll, I'll angel invest. [laughs] [laughs]

93. 15:41

    Thank you. Yes.

94. 15:45

    So what are some common trends with ideas that maybe you thought were good but didn't resonate with people and the reverse? Where have the surprises been?

95. 15:53

    Yeah, that's a really good question. Um, where have the surprises been? It's, it's been a little bit random, honestly. Um, I think,

96. 16:03

    I, I think one insight I got is, like, building apps where people can very easily share what they create tend to do way, way better. Um, and so now I try to think about that, like building in that viral loop into my apps on, like, if you can generate an image, I wanna make it really easy to

97. 16:18

    share that image or, and share it in with, like, a really nice OG image. Um, and so that's one insight I've had is like, you know, apps that have that viral loop of, of like, oh, this is really cool, and I can share this with a friend, and they can go on it and try it out, and

98. 16:30

    they can go on it. Um, and so that, that's something that I've incorporated a little bit more. Yeah.

99. 16:36

    Yes.

100. 16:37

     I love that you're making all of this, like, you know, it's all free so far. It's not monetized. But how are you paying for, like, all the compute and the model calls yourself?

101. 16:44

     Like, are you funding this?

102. 16:45

     Yeah, great question. So, um, first off, you know, I, I work at Together AI. We sponsor all of the compute in terms of the AI models. Uh, but then I'll also partner with other AI companies, like Neon gives me a free database to use their database and, and, like, you know, all, like Clark gives me a free

103. 17:00

     account for authentication. And, uh, uh, and the reason for that is 'cause it's, it's, it's all open source. So it's, it's in their best interest to be in that open source project and, and to, for, for people to use it.

104. 17:09

     Um, and so that, that's also, uh, like a good piece of feedback I have for people is like, if you wanna just tinker with apps and build them, a lot of people are like, "Oh, but I, I can't launch an app because AI models are too expensive," or, "This thing's too expensive."

105. 17:20

     Um, more often than not, if you launch something and you reach out to these companies and you say, "Hey, like I'm building this, this open source app, can you please give me some credits?"

106. 17:28

     More often than not, they will. Uh, and so that's generally how I do it.

107. 17:31

     Awesome. Thank you.

108. 17:31

     Yeah. Yes.

109. 17:33

     Yeah. Um, I mean, thank you for this talk. It was, it was really great. Um, I just had a question because you talk about how frequently you build these apps, like every, every week you said for, for a decent amount of time.

110. 17:44

     Um, and you also talked about, you know, using the latest and greatest in AI models. With them changing so quick and with you having so many apps out there, do you ever struggle with, like, going back through and, and, like, changing the models that you've used for some of these apps, and, and how do you deal with

111. 17:59

     that?

112. 18:00

     Yeah, it's a great question. Um, I, I, I think, like, a really cool, um, a really cool thing is the fact that you can do that. You can just build an app with an AI model, and then a better AI model comes out three months later, and you can go, and a lot of the time it's like

113. 18:13

     a one-line change of like, "Let me update this model," and the app just gets way better, or it just unlocks new things. Uh, and so that's something I do frequently, where I'll, I'll go back and I'll, like, even, like, relaunch an existing app with a new AI model or add a, a, a tiny feature to it.

114. 18:27

     Um, and so, um, yeah, I, I think that's kinda the superpower of, like, building with AI is the fact that you can just kinda replace these AI models.

115. 18:34

     Thank you.

116. 18:34

     Yeah. All right. Awesome. Thank you all so much for coming. Appreciate it. [applauds] [upbeat music]