No Degree, No Problem. Becoming a SWE, raising funds, succeeding in SF with no degree | Sam Hall

Sam Hall: I had this goal I'm gonna go to Stanford, I'm gonna go to Berkeley, I I'm gonna go to the West Coast and I I unfortunately did not get in. I will probably fail any type of coding interview you give me, but I can actually make things. And he texts me back like immediately after the call and he goes, It doesn't make any sense for you to do the technical interview, just fly out here. I'd quit my job. And so my biggest expenses were like rent, groceries, Google Cloud bill. And I was only spending $300 on this Google Cloud bill. I had roughly a month of rent left in my bank and it was two grand and rent was like $1,800. And so there was a hectic week after that getting the round closed, it was it was a roughly one and a half million dollar pre-seed round we called it and I I was like, I'm the king of the world, like I have I had this idea and now I have money. I really quickly figured out that getting the money is the easiest part because when you don't have money, the only thing to do is get money. And once you have the money, now you have five things you have to do all at the same time and you don't know how to do any of them.
Arjay McCandless: Sam, welcome to the pod. Thrill to have you on.
Sam Hall: Yeah, what's up RJ? I appreciate you having me.
Arjay McCandless: Not too much. So for a little bit of context, Sam and I are good friends. I was introduced to Sam through my fiancee. Guy is an absolute killer, moved to San Francisco with no degree, landed multiple six figure jobs, now is CEO, runs his own company. he's raised millions. And yeah, I want to talk about kind of his whole arc and hopefully you guys can learn from him. So I kind of want to start with Sam coming from like literally middle of nowhere, Midwest. I'm actually getting married like right near Sam's sometime in a week. But how did you kind of first get into tech, get into engineering, and then transition that into your current success?
Sam Hall: Yeah, I Basically, I spent a lot of time on the internet and and it it was sort of this like really beneficial thing where I actually didn't get the internet until I was like 11 years old and we were like third grade. And so it was sort of this thing where I was like so excited to get it that I once we had it I felt like I had to use it. And so I would just I was like a very like curious kid and grew up sort of in the country where there wasn't a lot going on, so I just go to the internet for everything and sort of explore the world through the internet. And more specifically, how I got into technology was I just like really loved watching. Like a lot of like sci-fi like space movies, and I was I want to like an aerospace engineer. And then I found out it's like, hey, turns out if you want to build any type of like drone or model rocket, that costs money, and they they explode because usually you crash the first few. And my parents did not like giving me the credit card to put on the internet, and so I was like, Okay, well, I can build a lot of things with software, and so I so I started like learning how to like write code because when you destroy code, you can just change a thing and recompile it, and now it works. And does it cost you anything? And so So I saved up, I I mowed a bunch of lawns, I worked at in a kitchen at like a local like sports bar. I worked there illegally actually because I wasn't supposed to be like using knives in the deep fryer when I was 14. But I used that money. I saved up for like a year and I bought like this pretty high spec PC and I learned like CAD, I learned how to like write code very basically and and because our our high school like like RJ said, we didn't have, you know, we were in the middle of nowhere, there weren't like computer science classes or or there were computer science classes. But I was you they added them like w while I was graduating. And so a lot of the learning was just me being online and and then kind of naturally I kind of got exposed to the more startup y world through this where I was seeing, you know, every time you go to a tutorial, it kind of links into some company that that builds like a product that you're using and then you kind of start learning about everything in the in Silicon Valley through that. And so I just kind of got fascinated with that and followed a bunch of people on Twitter and just lurked on Twitter and DM'd people really long rambly messages and people people really like to help when you seem very earnest. So people just kept kind of like entertaining me and it just kept going.
Arjay McCandless: What was your kind of first most memorable project you built?
Sam Hall: Well, it was it was a little ridiculous, but the first thing that I wanted to build 'cause So I was probably fifteen when I started, maybe fourteen when I started coding, so was like twenty sixteen, twenty seventeen. Like twenty sixteen and twenty eighteen was the the area. And this was when all like the cool reinforcement learning stuff that OpenAI was doing, like like the the OpenAI gym, like that product that they had sorta put out as like a a suite for people to mess around with. I wanna do that type of thing because I thought it was super interesting, and I was like, shit, I'm not smart enough to to be doing any of this and so I took a step back and But I really was interested, because I was still interested in like the flight controller and the drone stuff. And and so the one thing I could figure out was how convolutional neural networks worked. and at like a very high level, I did like MNIST, and then I was like, okay, I want to build a a model rocket with a camera in the nose cone that will then like point at a balloon. And and I was like, this is what I have to build. I did not succeed at that fully, but I bec because it turns out there's a lot of difficulty like pointing a c like point like getting a a a model rocket to point straight up and I kinda got stuck on like the Kalman filter section of that and so but I was asking people online and they were like dude you can't You're like actually what you're building is like I know it's like small, but it's technically illegal. You can't put like a guidance system on something like flies through the air like that. And and so people kinda I was like, okay, maybe it is best if the Kalman filter sort of stopped me from messing around with this. Because I was asking a lot of questions on the internet that probably indicated to to somebody that I was doing some weird stuff. but it wasn't dramatic at all. But it did lead me into working at like this this like drone startup and and and it was actually like this kid in college who was trying to retrofit these DGI drones to become delivery drones because basically you can use the DGI SDK to control the drone. So he was building an app around that. And it was actually very cool. You could basically take your like DGI drone and take like get like the starter kit from them and you could get it to go do deliveries. But there was this problem with the drones, which was that you're not allowed to like Take your eyes off the drone. Even if you're flying with a controller and you have a stream, you still have to keep your eyes on the drone. And so the drone will get in the area of the landing pad, but then the last sort of like the last descent is difficult because it has to drop below the tree line. And so he was like, Sam, can you figure out how to like detect the landing pad and land the drone on it? And I was like, Ooh, like, that might be pretty tough. But we it was actually, we I used like the Apple, I forget what they call it now, but there's like an ML kit. For Apple, where you can use like the you only look once model, and I think at the time it was like V3 or something. And so I used trained that model, did the landing pad detection, and then like wrote the code to like land the drone. And once I did that, I thought I was like the king of the world. And so that kind of just like led me into all these other things I worked on. But that was the first big one.
Arjay McCandless: And this was your first job.
Sam Hall: Yeah, except they didn't pay me. they he sent me a drone and I 'Cause I needed a drone to test. And so I was stoked. I was like, I got this two thousand dollar drone and but I didn't have any money and I couldn't sell the drone. And so so I was like, hey, is it okay if I start like doing like drone photography and videography to make money? So I made money that way and then and then I you know, I was just happy to have a drone.
Arjay McCandless: And how did you get connected with this person? I know you mentioned X, forums, was it through that?
Sam Hall: Well, actually, this is exactly the tutorial thing I talked about whereas he he was like s I I I think he went to like some university in Florida. He's actually a super sweet guy, but he went to some university in Florida and he would put out a lot of c t like coding tutorials that never got that big. But he put out this one called it was it was like a Spotify API wrapper in Python and it was called Spotify. And at the time I didn't really know how like any web development worked. I mean I didn't really know how any development worked, but I could like write basic like scripts that would run on my computer. And and so I was like learning about the Spotify library, and he was the only one that had tutorials on it. And so I I actually reached out to him in middle school and was like, hey, can you teach me how to use this? And he was super sweet, and he did. And then we caught up like a year later, and he was like, Hey, are you any good at this yet? Like, can you help out like my startup? And again, nobody was getting paid. He was in college, but I still thought that was like the coolest thing in the world because I was like, wow, like there's people actually building stuff. yeah, it was it was a lot of fun.
Arjay McCandless: think we'll into this more when talk about raising, but there's definitely something to be said about the advantage you have as a young person. Like people want to help young, ambitious people. Like they love that.
Sam Hall: Mm-hmm.
Arjay McCandless: And this is a perfect example of you like being interested and being also useful to him a year later, you know? Cool. So you
Sam Hall: Yeah, yeah. I think
Arjay McCandless: you land this job. Sorry, go ahead.
Sam Hall: I I think there's sorry, I think there's like a good point here which is like it it seems kind of spooky to like reach out to random people and and it's very difficult because, you know, when you're young and you're early you don't have you don't feel like you like a lot to offer somebody. but but most people they just appreciate it when you have like a very when like it's clear that you're reaching out to somebody because you've been like a really big fan of the things they've been putting out or like you really appreciate the work that they're doing. So a lot of people just like, you know, it's it is even even like relatively Okay. You know, there are these people who will get like 15,000 like followers on Twitter and you and like you know when I'm in high school, I'm like, wow, that's that's like a ton of followers, like they probably get DMs all the time. But actually, like a lot of times these people don't get like that much appreciation for like the work that they do. And so if you come up and you send them maybe maybe you're not very good at writing succinct emails, but maybe you send them a three-paragraph email and you get really excited about what they're doing, they love that. It's like the one time that somebody's actually being like, Wow, I really like the thing you're doing, and and so people love it when you do that. and they wanna help you a lot when when they see like that you're earnest and excited about just kind of the world that that they're in. And and there's a lot of pay it forward attitude, I think.
Arjay McCandless: I also do think that a lot of people I meet will DM, you know, the head of product at Claude or OpenAI and they won't get a response. And it's like, well, okay, those are people who are actually getting 15,000 DMs. So
Sam Hall: Yeah.
Arjay McCandless: scope your DMs appropriately. And then as you become more, you know, influential, you can kind of work your way up the ladder of cold outreach. But yeah, I think those like fifteen hundred people who are in your niche that you want to be in, perfect. So you
Sam Hall: Mm-hmm. Definitely.
Arjay McCandless: have the unpaid drone job. Now how do you land your first paying job and move to California and all that?
Sam Hall: Yeah, there there is a direct lineage here which was that the unpaying drone job got me paid like videography, like photography jobs and it was basically I benefited heavily from from being in high school and not needing that much money. So when I'd go do like videography, I'd do it for Realtors. Realtors was like the smartest. Like this is where I learned like a little bit about sales and and like how to find a customer because we do drone photography and videography, and I'd call all these resorts and they'd be like, Yeah, I I'll be interested, and I'd charge like a hundred dollars, which is like pretty we were putting out some pretty good like videos and sending them to them. but they're not repeating, you know. Once you have like a video of your resort, they're not gonna buy hire you. you again because you did a good job the first time. But realtors are perfect because they will keep hiring you, plus they're making money directly off of your videos. And so I would charge really low prices and it pissed I think it pissed some of the people off that were actually trying to like make a living off of that. But but the problem was was I was taking all these drone videos and I was trying to take like I was trying to build up a little bit of like a portfolio and I couldn't pay for Google Drive because I didn't have a credit card and my parents didn't want me to to put like a recurring subscription on a credit card that they had because they knew I'd forget about it at some point. So I was like, well, Google Drive is Just a computer on the internet, like why can't my my computer's on the internet? Why can't I build a local version of Google Drive so I can do peer-to-peer file sharing with one of my friends who I was doing this stuff with? And very naively, I thought this was the smartest thing in the world. And I I love to nerd out about this a little bit still because I still think there's like there there was some novelty to this this project I was working on, but I basically was reading about this thing that had recently come out and added to like the the like browser like API like like a like in Chromium and and most browsers. And it was called the file system access API. And the file system access API was basically a thing where your Chrome inst like your your Chrome instance and therefore as a developer of a web app you could like have a user upload files to your web app, but all it did was create like a a pointer and say like hey here's like a handle into your file system and you could do like folders, you could do individual files. And so the thing that I built basically would say I could take like you know I could have a terabyte of of video footage, I could drag and drop it into this window and it wouldn't transfer anything, it would stay on my computer and then we could show previews so that if somebody like wanted to do like the connect over a web WebSocket or connect over at WebRTC, they could go and click on the the previewed file that they had seen and then it would transfer. And the nice thing about this was I like, you know, our internet, you know, we the one we got in third grade was pretty bad. And so the nice thing was somebody didn't have to wait for me to fully upload something in Google Drive to get it. And I thought this was like the smartest thing in the world. I was like, we are gonna kill Google. I didn't realize that nobody else, I'm the only person probably like in the world that that like gives a shit about like having to pay for like Google Drive. And but But I thought this was very cool and I was very proud of it. I I still think it was actually pretty cool. This is also before like clawed code. So I actually had to like learn everything. And so I was like, wow, this is like the first time I built like a recursive React function. And I was like, I I remember having to type into Google what is like how do you make a function call a function like its own like itself? And then it was like this is called recursion. And yes, React like functional components can like use recursion. Because I had to rebuild like a file system tree. And and so this is like how I learned everything. And and Basically I had this, I was super proud of it. And it was at the same time I was graduating high school and I I had this sort of like goal of like I'm gonna go to Stanford, I'm gonna go to Berkeley, I I'm gonna go to like the West Coast and I I unfortunately did not get in. I was very like down, I was feeling very down and and I was like, man, what do I do? Like I can't like I've been sad for like
Arjay McCandless: Did you get in anywhere?
Sam Hall: I basically got into San Diego and and and and I sort of this is like another thing is I sort of had like the hubris where where I was like I I like I would look up like the average scores for these schools. I was like, okay, I'm in there. I'm good. And I didn't realize that everybody
Arjay McCandless: yeah. No.
Sam Hall: applying, every single person applying was was in those ranges. And even of that pe of that like like group of people applying, four percent was was was of that sample. And I was like, no, like this is terrible. And so I didn't apply to like any state schools. I did like I was scrambling because I also didn't want to go to San Diego that bad to pent spend like 40 grand out of state. And so I just was like, man, maybe maybe this is like a good opportunity for me to like go and take this this peer-to-peer file sharing app like to the limit. So so again, I go on Twitter, I haven't tweeted anything at this point. I I have no followers, but there was this angel investor that I I watched like tweet, and he goes, or I DM him and I was like, I I this is the other important thing, is I actually benefited heavily because I treated this DM with like a lot of I put a lot of effort into messaging him. I listened to a podcast episode he's on. I went and like learned everything about his company. I learned about the types of companies that he angel invested in before and I I sent him a demo video. I sent him like a very long email, and he basically responds like, very cool. This is a cool like project. It'll never be a product. and he goes, But if you can build this, you could you could come work at my company. And his company was this company called Socra, and it was this like FinTech style. Like you you basically would go there if you're like a VC or something, and you'd get like research and reports about the private markets. and it was like a team of six people, there's one one other software engineer, and I was like, this is amazing. And and you know they offered me like sixty thousand dollars a year as like I was like, holy I made it. Like this we're doing really good. And and so I was very excited about this and I and my parents were like, you know, my parents were like, wow that's really a substantial amount of money. you know, actually I guess if this is real, like yeah, I guess take some time, like don't go to school. But maybe you'll go next year. Like maybe, maybe this will just be a gap. because my parents kind of they didn't push me that hard, but they did want me to go. And so yeah, that was kind of how I got in the first job. And it was like I started like a week after graduation and it kind of happened out of nowhere. It wasn't planned or anything.
Arjay McCandless: So you send one cold DM and end up with a job. That was seriously it?
Sam Hall: Yeah, it was basically I I think I sent an it was actually an email. I don't think I DM'd him, but I sent him an email and I like referenced the podcast he was in. I was like yeah, I would basically always just if I wanted to talk to somebody and I wanted like to a make an ask, I'd always go try to find like one thing to at least show I did the work of like why am I asking you? Like like why why is this not like an email I could have sent to a thousand people? Like how do you make it like very personalized? And and he just approached appreciate I think it like he just appreciated like, okay, like here's like this eighteen year old that Did not write a very good email, but he definitely has a lot of energy. And and that that kind of that was kind of everything. Everything just kept going off of like, wow, okay, like I have no clue if this kid's gonna be good at this or not, but you know, he's definitely got the energy for it. And so yeah, it was just one email. We got on a call, he was like, Yeah, talk to the the other engineer. The other engineer just asked me some questions, because I didn't know how to do leak code, I didn't know like any of the stuff that you talk about on your channel. Like even when I watch your channel now, I'm still learning stuff, I'm like I'm like, well I didn't know about that at all. I should probably know that. And And so it was very much he was like, Yeah, tell me about this project that you showed Walter, who was the CEO and and so I was just going through it and he's like, Yeah, you know, you s you seem like you know enough and and and that was it. It was very like and then and then yeah, I just kinda had to like once I started I had to like really like focus and learn a lot very quickly and be able to like take on a lot of ownership. And so I I think it kinda just comes down to like, you know, if you if you don't have the experience you have to replace it with a lot of energy and a lot of hard work and then over time that starts to invert where now people just wanna like understand your expertise. But I didn't have any expertise so I just had to like brute force my way my way through
Arjay McCandless: I think this was probably a very good cold email. Even though you say it wasn't, I'm sure it was excellent. yeah, you you should find it. I would I would love to to put it on screen here. But no, I mean
Sam Hall: We'll have to find it, dude. It's so embarrassing.
Arjay McCandless: I get a lot of cold DMs and cold emails. And I think it's gotten harder because a lot of the emails I get now are very clearly somebody told Claude, transcribe five of this guy's YouTube videos, write an email that seems personal and send it. And please don't do this. It it it's obvious and it doesn't feel personal at all. It just feels like you wanted me to think you spent time, but I actually didn't. And this is like I don't know. I don't think that mass cold outreach is the way. I think the way you did it is perfect. Like actually pick something you're deeply interested in and then put in the work and send the email. I love that story. So you're at this company. How many more jobs did you have before deciding to start your own thing?
Sam Hall: It was three jobs, only two that made it to the resume, because the third I I quit pretty quickly. but yeah, from there, the the sort of manager that I referred to that that that interviewed me three months after I joined, he left. And this was like a very scary thing for me because I was not like, you know, I definitely not a good engineer. not even like not a very good engineer. I was just not a good one. and so he was leaving and I had to take on and again it's like a web app. It wasn't necessarily the most sophisticated thing, but but there's still so many things that confuse me about like the structure of everything and and so I basically and again before Claude Code, so I couldn't just like have Claude help me understand the architecture. It was very much like, hey, get on a call with me, like show me all these things. And so he left and my my boss called me and was like, Hey Sam, like you know So-and-so is leaving, we're not going to hire anybody else. Like, I think you can handle it. Like, you need to step up and be that person. And this was like, it was very, I was very excited about it. Like, I at least like they had that they had faith in that. And then also, but it was also very terrifying because you know, it was it was like, I don't know what I'm doing, I'm gonna mess up a lot. And so I stayed there for another year or so, and and mostly it was because at a certain point I started to really appreciate. The ownership and the ownership gave me so much confidence in myself where it was like I started off being terrified of like you know I'd get texted in the morning like hey can we hop on a call? I'd like terrified to get on the call. I was always like nervous about deadlines and things like this, and eventually I was like, I know the most I know more about this than anybody else at the company, and and that's like it actually which started off as a very like scary thing, became like a very freeing thing, and it made me like really appreciate my own capabilities when they were like really tested. And so I felt really good about like towards the end of this job I felt really good about sort of my like own personal improvement. But I did definitely want to like you know, I I was still like nineteen and I definitely should not have been the only engineer somewhere and and so I saw it was probably a good idea to surround myself and go to a company where there are like experienced engineers that I could learn from. And so That manager, he had joined a company in Palo Alto that was roughly like 40 people and and there's about a team of 20 engineers. He goes, Sam, like you should come out here, like I'll refer you, and then I think you'd fit in really well. And so I was like, Okay, I get to move to California, I get to I get to be around like many more people and it's it's a startup that's raised like significant money. And and so they they flew me out for a work trial, it was like three days and the CTO who actually became like a r a mentor of mine, the CTO I got on a call with him and I was like, Look, I I also this is the thing I think is important is like if you if you're very earnest and you explain like you just just be like transparent. You don't like you know, when you're young you don't have a lot of leverage. Like people just appreciate like transparency and like like earnestness. So I'm on the call with the CTO and I was like, Look, we're c it's kinda like a like a vibe check call and I was like, look, I don't know like and I promise you like I'm gonna be the worst like technical engineer. I will probably fail like any type of like coding interview you give me, but I can actually make things. I might spend a little bit more time on it, but I really like pride myself in like in being able to build the thing and and I'll spend more time to make sure I get caught up and I try to learn very quickly. And and he goes okay that sounds good. I think we'll still have you do like the technical interview and like you can just see But we probably will have you come out for an aunt like a a a work trial. And he texts me back like immediately after the call and he goes, It doesn't make any sense for you to do the technical interview, just fly out here. And so so I was so glad because I was like, no, I don't wanna
Arjay McCandless: Yeah.
Sam Hall: get stopped now. So I fly out, I'm in Palo Alto, and they basically just had me work on a feature for three days, which is so fun because you kind of like inherit this like big code base and you're like they're like, Okay, now go make this specific thing. You get to meet everybody, ask them all questions and And yeah, I I didn't finish the full feature for the thing, but I I had made significant enough progress and I had talked to enough people on the team and and like gotten like gotten like learned from them and why like they had built things that they did and everyone liked me enough where they decided to hire me and so I I moved out moved out to California and that was sort of like the first like startup kind of job. Because the first company was like was definitely ran more of more more like a It was almost more like a content style company. It wasn't necessarily like a heavy VC backed like startup, like Silicon Valley like flavored startup.
Arjay McCandless: So I wanna touch on the work trial thing quick because I'd love work trials. I think that's probably the only way to hire right now. Whenever they come up, a bunch of people get butthurt about you expect me to spend three days interviewing for you and then you might not hire me. Like I'm curious how you saw the work trial, something you were excited for, is it something where you were like, I'm gonna give you three days of unpaid labor? Like, how did you see it? And
Sam Hall: Yeah.
Arjay McCandless: do you still believe in work trials?
Sam Hall: I well A actually they paid me and so that helps a lot because it it sort of like I actually didn't care. I I would have I I wouldn't have cared if they paid me or not, but I think that's like a good thing if you're studying companies pay like people for work trials because then at least like A it it kind of you already have to sign some type of contract that says anything that they work like do is your IP. But also like the if you're doing a work trial, it's very unlikely that the company unless you're like one of the best engineers, in which case they probably wouldn't make you do a work trial. Like if you were like really like coveted of of like a hire, they probably won't like have you do a work trial. They'll probably just like like screen you and then hire you pretty quickly. But I think Most likely the company's not gonna use like the thing that you made. It it might even be a thing that they made up as a problem. because like in three days you're onboarding into a totally new code base. There's it's very unlikely that they're gonna be like, this perfect, let's ship it. So I wasn't worried about that. I was getting paid. And also I felt it's like, man, if I'm doing a work trial, it's also my opportunity to like gauge if I think this would be a good environment for me to be in. It's also like I I was like I said, I wasn't like the best engineer, so I used it to like highlight my other like my other strong points. It's like I'm very curious, I really like learning from people, and and so if you can like find a c like a team that has a culture that they like to teach people and they like to like sort of foster that curiosity, then you win. They they they get to like see that you are really excited. So even if maybe you're not the best engineer now, they can sort of bet on your growth. And that's like the big thing is a lot of companies are willing to take bets on somebody who can grow because Like every single person that goes off to do really like impressive things or becomes really good, like the only way that they can hire those people, like like early in their career is that they make a really big bet. Like there's not if if if by the time you are like you are you like if you're like deeper in your career and you have all these credentials, like it's not a bet anymore. And now now companies are gonna hire you, like men you'll have many offers. Whereas for me it was like I had to convince them that I was a bet worth like taking and and they were like, okay, well this is a good opportunity. We could potentially have a really good engineer here that we just have to like build into that engineer. And so I think work trials are like a super good way to like expose like to to basically prove that and you get more opportunities to to highlight yourself. from like you know I we ran work trials when we were interviewing people and it was like a really good way to also see like like does the team enjoy like maybe they're a great engineer and especially now where it's more of like they have to be the best engineer like they have to be very thoughtful and smart and hardworking but they maybe don't have to be the best like person to like they don't have to write the best code or do leak codes necessarily but like do I enjoy working with this person do I do I like want to come spend more time at the office because that's like a real killer if your team is small if somebody like doesn't get along with somebody else else on the team and now nobody really wants to be there. You like you're tanking like productivity. So it's like you just want to make sure that everybody like loves the the team that they're on and and feels like fueled up to be in the office.
Arjay McCandless: Yeah, I want to double down on this. This is something my CEO at Maven said when he was hiring and his he was a mentor to me, excellent CEO. He basically said at Maven they hire for high slope. And I it's just a simple way to put it. Like you're hiring for somebody that is maybe not at the apex of their career, but they're have a great trajectory. And this is really nice because as a candidate, I was hired to work in a marketing role with zero marketing on my resume. And he was just like, Yeah, I feel like you can do this. And I was like, I do too. And that's, you know, you have to present yourself as somebody who is high slope. And that is like kind of what you've been getting out, I think, through a lot of this, just like being really earnest, being trying to prove that you're willing to do the work. And I think that's like super important. Okay. So work trials, great. I want to talk about your company. So at what point do you decide, hey, working for people's school, but I want to do my own thing. And how does it kind of happen?
Sam Hall: Yeah, I it it wasn't exactly very linear where there was like a thought in my head was like I was like man I I really want to start a company. What what actually happened was so I was working at this company in Palo Alto and I and and and no no like shade to that product but I I wasn't like personally like like a customer of that product and I I so I therefore wasn't you know like the most you know I I wasn't like the most in love with the work I was doing every day. But it was very good for me to learn and become a better engineer. But then you know as you spend time in the environment like especially San Francisco It's like you step outside and there's a self-driving car, and then you you're just like it's crazy, you see all these crazy things happening, and so I was I was thinking, man, like what do I like what do I want to work on? Like what do I think is really interesting? And so that was the first thought. And so I then I at this point I had spent enough time in the Bay Area and and I just had met enough people through like events, and my neighbors happened to like have a startup that was actually like relatively big. and and so it was like a lot of it was a lot of just like very happenstance. I had met a lot of people and so I started asking them. I was like, I'm interested in this, this, and this. Like, do you know anybody I can talk to that that knows about those things? And so they started making introductions to for me. And I was just really interested in like really just trying to explore what I thought was like exciting to me. And I got introduced to the CEO of a company called Lindy, and and they were building like this, it was kind of like AI, like Zapier is like sort of for enterprises. it was one of like the first agent products, and this was like genuinely, I'm I I genuinely mean this. It was I had never heard the term agent before, and now everybody uses it, but they were saying agent, and I was trying to explain it even to like my friends, and they're like, Okay, what does like agent mean? And I was like, Well, if you imagine like an LLM is like a talking head, they're sort of giving it the hands, they're like letting it actually do things in in in the real world. and And so I was like, this is super interesting to me, and so I want to be a part of it. And so I joined I got introduced to the founder and the founder basically said, Hey, we're not hiring anybody right now. But if you're okay with that, if you're willing to like waste your time going through the process, like there might, there's a slim chance that that we will like hire you. And I went through the pro I also had time, you know, I had time and I was I was trying to get something that that maybe I didn't deserve that early in my career. Like I I don't think, you know, I was very lucky to be in a position where I could say, I'm really interested. in this, how do I start working on it? And so I was like, okay, I'll waste some time like trying to like explore this. Like you know, never you never know what happens. And they ended up hiring me. and And I was excited, but I got kind of stuck on they were like, okay, you know, you're new, like, go fix bugs, go fix these like CSS things. And I was like, I had explicitly told, told the CEO when I joined, I was like, look, I know it's a startup, I know I gotta work on the things that are most important, and like the the highest leverage in the moment, but I really wanna work on specific like the infrastructure, like the agent, like the agent harness, like this type of thing. And I was like, I will leave if I don't like think. It's I'm working on like exciting thing. And I was like, I under like you, but just know that. Like you don't have to hire me. But just know that this is my intention of coming in here. And and so I was very upfront. And then he was like, no, it's okay. And I still ended up working on CSS things. And so so I was like, I don't really know much about CSS. But but I I also started having my own ideas after kind of seeing their product. And so while I'm working there, my mind is kind of wandering. And I started thinking about like what is the most like consumer version of like this agentic product a product where it's like I can talk to my computer and now it can start to do things for me and it can start to like like like convert my like high level instructions into very like direct like interaction with the rest of like my computer or my device or whatever. And so so I was like Well, I feel like this has to live on the phone and I feel like it needs to sort of inherit all the context that your phone already has. And if you have all that, then your instructions can be super vague and and you can ask it to do so many different things and it can use all the context to sort of understand what your intentions are, predict what you'd want to go do with all the apps that are already installed, and then it can even go and like interact with the apps and and start doing things on your behalf. And so I I basically was like, I think it's possible, I'm gonna quit and just mess around with that for a little while. And I was also fortunate enough enough at that point to have had saved up some money working at the startups to be able to go and like spend six months just messing around, like trying to figure things out.
Arjay McCandless: Remember this. It like you make it sound a lot safer than I felt like it was. But I was working
Sam Hall: Right.
Arjay McCandless: at Amazon at the time. I'm like, this guy's quitting his job with no plan and like not that much money in San Francisco. Okay. It was it
Sam Hall: Yeah.
Arjay McCandless: was a risk on your part for sure. but so you okay, so you start building, then one.
Sam Hall: So it was basically this was like probably the most stressful like period because I was excited about it. I I I just but I again like I thought maybe this is a company, but I was also keeping my eye open, like maybe there's another company that that's already a little bit more established. Well, a little bit more established than nothing at all. So you know this there's a lot of options. I kept my eyes open trying to see like does anybody kind of have the same thesis as me? But then as I continued working on On it, I got more and more opinionated about how it should work. And and to take a step back, specifically the sort of entry point that I had was like I started with this criteria. I was like, it has to be like an almost out-of-the-box experience, it has to be able to see everything that could be important to you, and I don't want you to have to like do this like very insane setup. And and so my sort of intuition was like, Okay, well, what is the most like personal device that sort of has the most context in my life? And it's like, well, my phone, I have my phone everywhere I've used a phone for like a decade now and And so I was like, okay, but I can't build an app and see the data from another app. So like my agent can't use another app if it exists as an app. And I was like, well, if you live below the apps, like if you live at the operating system level layer, you can you are the thing doing the sandboxing. So why don't you go and like leverage that and build this sort of cohesive layer underneath where agents can operate within? And I was very excited about this and I but I had no clue what I was doing. So it was like I Android is open source, and so I started there. So there's like the Android. open source project. I turns out I found this out after I quit. Every all this information I found out after I quit, but Android's build system is Linux-based. you need at least 64 gigabytes of RAM for the build system because the source code for Android is so big. And so basically I had a lot of things to figure out. I rented a computer on Google Cloud. And again, I'd quit my job. And so my biggest expenses were like rent, groceries, Google Cloud bill. And I was only spending like $300 on this Google Cloud bill. And and so I was like hacking on this and I messaged one of my friends. I was like, yo, dude, like I'm starting to work on this, except this Google Cloud bill is so high. And he goes, I'll ship you like this gaming computer I have. And then he goes to FedEx and it's like it's $500. He goes, You should just fly here and come get it. So I get this computer. I like set up this build system on the computer and I'm kind of like tinkering around. I'm like using emulators to like like flash like these emulated phones and and we start to like I start to understand how Android works. I start to see all the data that we can build at this like sort of central OS location and and I was like wow this is actually like I'm starting to see it now it's not just in my head And I put a video out on Twitter and I this at this point, it's like six months later, and I was like, I love this. Like I haven't found anybody like doing the same thing we're doing. I haven't like found a team I can go join and I really need to keep doing this. And I really like I'm now very strongly opinionated about how it should exist. And so basically I was just like I fell in love with it and I just kind of asked myself like how do I keep doing this? And so I had like roughly a month of rent left in my bank and it was like Yeah, like two grand and rent was like $1,800. And I was like, all right, we gotta figure this out quickly. So I posted a video to Twitter. And at the time I had like no Twitter followers, but I known enough people in person and they started sharing it and it it went it like it did pretty well for for having no followers. And at the same time I had met some friends in sort of the startup scene that were introducing me to investors and they had to introduce me to investors and investors were like, Yeah, let's talk in a couple weeks, like like let's do this, let's do that. And the like the Twitter video went out and enough people saw it, then those investors saw it and they're like, Can we talk tomorrow? Can we like, you can we Can we talk sooner? and so it kind of like started this sort of like rapid two-week period where I was talking to like 40 investors over two weeks and at the like at the end somebody's like here's like a term sheet and then another week goes by where all these other investors like you got a term sheet like you're not supposed to say you have a term sheet, like you can say you have a term sheet, you're not supposed say the terms, that type of thing. And but people find out, like the value is very small. And so there was like a hectic week after that getting the round closed, and then I just it was it was like a roughly like one and a half million dollar like C round or pre-seed round we called it and I I was like, I'm the king of the world, like I have I had this idea and now I have money. Like like I can't be stopped. And and then I really quickly figured out that getting the money is the easiest part because when you don't have money, the only thing to do is get money. And once you have the money, now you have five things you have to do all at the same time and you don't know how to do any of them. And So it kind of like, you know, that was sort of the the how it turned into like a real company. But it's very much just like motivated by by me saying like, wow, like this is something I really love. Like I how do I keep doing it? Like how do I how do I take it to the next level?
Arjay McCandless: Did you think about applying for accelerators and stuff like that?
Sam Hall: I did kind of skip over a little bit of a of a detail, which is that there's this thing in in San Francisco called South Park Commons. And South Park Commons is basically like the TLDR, is it's like it's kind of like a YC style accelerator. However, around that, they also have this thing, just it's like a membership, but you still have to apply. So their accelerator is called the Founder Fellowship, and I forget the exact deal terms, but they're actually like, I think they're competitive or slightly better than than the YC deal, except the founder fellowship. only accepts like I think like twelve teams every like half a year or or every quarter or whatever it is now. but the membership they accept like like like ten people like every month or twenty people every month. And so I applied to the fellowship And they're like, you know, maybe you should spend some time like finding a co founder and in this type of thing and and so maybe you should come apply to be a member. And, you know, I I was just working out of my apartment and I thought it'd be cool to to be part of this. And so I applied as a member, I got in and there I met quite a few people and that was I that was a very exciting experience because Everybody was very like empathetic to to the sort of struggles. they they were like empathetic to to like trying all these things and nothing working and and and so I met a lot of really like great people there. And that also helps a lot with the fundraise because all these people now are like introducing me to their people and yeah it was a it was a very good like thing to be a part of, but I didn't I didn't receive any money from them. But I definitely think the hardest thing to do is is when you're starting a company you should probably try to surround yourself with as many people as you can that you like you at least feel that like they they're like they're excited about you, they're excited about like kind of they're able to bring you up and you can share ideas with them and and it they become like really good friends 'cause they kind of like see you when you're doing the very worst. So so they they become like really close with you.
Arjay McCandless: Yeah. Okay. Let's get a little bit technical. I'd love to play a demo of your product here if you have one. I wanted to put together a quick video showing my phone running our custom version of Android, where we've modified the OS to make it so that agents can see everything across all of your apps. And that's only possible if you're actually modifying the underlying operating system. And what you're seeing here is my home screen, which is this sort of compiled view of all the things that I care about, and my my agents like determining this based on all the stuff happening inside of my phone. And I really felt that. You know, if you open up your phone, you just see a ton of app icons with these little red bubbles, that's not really like helpful to me. I don't think of my life in app icons. And so we wanted to build a home screen that shows you all the stuff that you know you have going on. And so here you can kind of see like updates about the people in my life and scheduled meetings. It suggests quick actions to go respond to them, set up, like you know, make sure I can make it to the coffee, message Kyle seeing when he wants to hop on, you know, a a call to get feedback. But I'll show you You know, I I could go through this a lot, but I I wanna keep this short. I'll show you the other thing you can do, which is you can search. Can you put together a summary of how much we spent in the last month at shipping phones? I I had a hard time getting that one out. But so here what it'll do is very similar to the to the home screen, it's gonna take that question and go search for that information across, you know, BRECS receipts. the FedE, my Gmail receipts, and any potential, you know, shipping labels that were sent and try to kind of merge that information and get an idea of of the phones we've been shipping out. It'll limit it to the last month. And and yeah, I really like this example because you know there's so much more data inside of our phones than we necessarily think. And as long as you have a system that can see it all and like reason about it, then you can actually do a lot with it. And so here it's breaking down that we spent 848 at DHL and 244 on FedEx. DHL is just so much more expensive. but it kind of breaks down all the the different shipments in chronological order. and now I can say something like, you know, now that this information is compiled, you know, maybe I want to do something with it. So I can say, can you find the average price for both services and send that finding to Garrett on Slack? Now, you know, it's not, you know, it's not the most novel thing to be able to message somebody on Slack, obviously. But what what we're kind of doing here is we're trying to sort of close the loop of once once the system's able to go compile all this information, can I go and do something with it? And so here it's controlling this pip window, which this is the actual app running on my phone and the agent's controlling it, and I can move this around, I can close it out like this. and and one cool thing that we do here is we try to go back and read. your messages with that person in the same conversation channel and clone your voice and and that way you know it doesn't sound like you're getting a message from this AI system. So you can see it it broke down the findings, sent it to him on Slack, and it jumps straight to our conversation here. and it kind of gives you a summary of what it did. And then I can actually I can I can swipe it down here, but I can also, you know, if if the agent's running a longer task, I can say, okay, you know, I don't really like what you're doing. I want to take over. And I can click this expand icon and it'll drop me into the app and the agent will stop running. That was obviously a really quick example, but I just wanted to you know I just wanted to make this quick and show you kind of what we can do. But yeah, love your feedback and any sort of thoughts or questions you have. So now we know what the product does. I'm curious you mentioned it operates at the operating system level. I'm kind of curious exactly how you designed it, you know, what you're willing to share around the specifics of the engineering behind it and especially working within the constraints of being on a phone.
Sam Hall: Definitely. So I'm open to sharing everything about it because I've talked about it a lot before. but basically, you know, the the underlying Thesis was to get as much data as possible. And so the first place, like the the ground zero of getting data is always like, okay, can we look at the databases? And there's something kind of unique about the mobile operating system, which is that there's a lot of data in the databases, just given the fact that a lot of developers try to optimize their apps even with like poor cellular connectivity. And so, even like Snapchat, for example, like you have to be able to like receive the message, and then you have to be able to open it even when your internet is spotted. or your or your cellular spody. So Snapchat will store that in a SQLite database, because that's sort of like the primitive database that Android's built around. And the second it hits that database, you know, if you're sitting at the operating system layer, you're like you're able to to read into that. You have to modify the Android is technically set up to have some like security enforced Linux policy that says like you shouldn't be able to access that data. But we just changed the policy and gave our like system or we gave our like system service We give our system service access to the to the data. And there's arguments against why maybe you shouldn't do that. like like like in a business technical perspective, but but we wanted to just see like, hey, what does the product feel like when when it has all this when it has all this context? So the first thing we did was we figured out that we could read all these databases, and then we had to say, okay, how do we then take any arbitrary app, scan its databases, and then normalize that into our centralized index because you can kind of like think about what we did for the phone as this like sort of similar to like a web index where we're trying to take all this data and and normalize it so it's easily queryable and And so we had to build these ingesters we called them, which would go and like scan the databases, look for any type of text that might be, you know, might contain some semantics or some relevance to your life. So you could think like text messages, Twitter DMs, Instagram DMs, even like the Whoop database would publish like your health data to it. so we were trying to figure out how can we like take all these signals and can we normalize them into into like a central place. So we started off with this thing called we called it like a generic ingester. and so we'd go like, we'd do a scan the database and we'd check, does this row or does this column, like we'd sample like an entire database, we'd look at a column and say, does this column, do any of the rows Or do do any of the column values for all of these rows contain like like s like human readable text basically? And once they did, once we like identified that they did, we would we would like kind of keep a record of that. We called it like a mapping. So we'd say like, okay, the Snapchat database on this on this or the Snapchat app with this database with this table and this column has like a value that we need to read from and continuously read from. So we'd discover that and we'd save that. And then we'd go through sorry, you
Arjay McCandless: And are you sorry, are you duplicating this data to your own systems or you saying like, Hey, this data's here, if you want more from Snapchat, look at it.
Sam Hall: So we duplicate the data to our system. initially this was not the plan, but I can briefly touch on that in a little bit. But so so we basically built, we we'd sort of extracted like human-readable text, we'd extract timestamps, and we'd try to also like try to do some like dynamic joins. Because, for example, if you know RJ sends me a text message, the database represents it as like his phone number and then the text message and then the timestamp that it was sent at, but his name and his contact is saved on. a different table that we have to join on. And so to make it generic we had to say like, okay, can we kind of like do all these joins and see like do do these values map? Like does this contain like a username field? So there's a lot of like almost hard-coded like regular expressions and and like hard coded like rules where it's like if this seems like a username field that we should join it on this like account ID field, things like this. And so there's a lot of work put into that side of it. And then we would take all of that data and move it into our database. And now the reason that we duplicate it, the reason we duplicate it from the databases is because most databases on the phone they act as a somewhat of a cache. You can expect them to like be rewritten. And and we would also even read cache caches. Like if you, for example, like a lot of apps started using GraphQL. Linear, for example, would use GraphQL and you'd use the Apollo client. And so the only data Data we could get from Linear was data that was put inside of the like GraphQL like Apollo client and stored locally in the cache, which was already a SQLite format. And so this would get reset all the time. And so we had to duplicate it so that we'd have this, like maintain this hor historical sort of like massive like database of all the things that you've seen through your phone. And and then, well, you know, once we had this database, we started indexing our own database. We started trying to say things like, can we extract subjects? Actually maybe I should take a maybe I should pick take a brief pause there because the this is where we get into like all of like how do you build a retrieval system, like what is like our
Arjay McCandless: Yeah.
Sam Hall: sort of philosophy on retrieval systems.
Arjay McCandless: Th yeah, no, that was super useful. high level question how did you learn this stuff, or was it more just kind of hacking?
Sam Hall: It was it was all hacking. It basically through trial and error. I would I would read a lot. I would I would go like I would basically try to go like deep if I ever ran into an issue, I'd be like, okay, let me try to like consume as much information on the internet as I can to to understand this. But it was really just like, okay, I made a simple version like I made a simple version of it, but it doesn't work for this case. How do I expand it for this case? So well now it's all really slow. I don't like how slow it is, how do I make it fast? So it was very much it was a very much like just trial and error, how come it doesn't do the thing that I wanted to do? And I kept refining it until it started to actually do that. And then Cloud Code came out and I just said okay, make it do the thing I want it to.
Arjay McCandless: Yeah. And how much of this stuff is happening on device versus on the cloud?
Sam Hall: Our our indexing and our like our scanning of the databases, that's in the cloud. Or not in the cloud, that's that's on the phone. The when it when a model has to go infer something based on the data, such as like, you know, the proactive home screen that we showed in the demo, or like if you ask the phone a question, that the model driving that and the model like kind of parsing over the results that it gets back, that's all in the cloud. That's using like Gemini, three and a half flash right now. the models are just not good
Arjay McCandless: And a half.
Sam Hall: enough yet. They're not good enough yet to to to run on device and get like a really reasonable result. In in fact they're very slow on device right now because like on device inference has not heavily been optimized and and so it just build a really bad experience.
Arjay McCandless: That makes sense. Okay. Let's get into retrieval systems. I personally think that retrieval systems are basically the bread and butter of most AI applications right now. So if you're interested in that stuff, like pay attention. We're gonna go pretty deep here. So lock in.
Sam Hall: Like Yeah, I I see a lot like I see a lot of these like posts where it's like rag is dead and it's like I think I think rag as a term is is a little bit loaded. I I think with with the sort of new AI wave we've like we like keep attributing new terms to systems that have existed since we've been building like software. For example, like there there's this There's this book, I think it's called Building Recommendation Systems in JAX or Recommendations and Retrieval S Systems in JAX. But it's authored by this guy named Brian Biscoff and he's awesome. I I met him a few times. He's he's he's actually testing one of our phones. But he he argues that's like all that rag is, is it's just a retrieval system it's just it's just a like a recommendation system for what goes into the model. It's no different than like like, you know, like when I when I go and like search for something on on Google and they have to recommend me an ad based on my search and based on my history, it's just they're recommending me and and and putting in like an ad. And then this is you trying to build something that's like, okay, how do we find the most relevant context to recommend to the model? And so rag kinda became a thing because people are like, there's something novel about the fact that now we have these embeddings that come from these like the hidden layer that we chop off of a of a model and then we represent them as numbers and now we can do like, you know, basically like trigonometry, like cosine similarity to say like these are similar. And and that's like that's like a most I think eighty any rag system or recommendation system that you see in production is very unlikely just doing cosine similarity. And so it kind of already feels like it it kind of it kind of never seemed like a really distinct thing to me. And we never tried to treat it as like a one-shot like silver bullet to the recommendation or retrieval problem. And and so the way we kind of built our system was we had to build around SQLite. It's like very heavily supported in Android. And so we built our own like extensions where you could do cosine similarity and it was relatively fast. And so that was like we added that, but then also we just leveraged like the relational like nature of of these of these databases. Like for example, like I can I can kind of break down the the main things that we would try to pull out to feed to our to our model. But basically like our index, what we wanted it to do is be able to capture semantic information, so that's where like the cosine comes, like the the embeddings come in. You wanted to be able capture temporal information. You want to understand sort of the the who, like the nouns that become like very important in your life. So it's like we have to pull out named entities and do this type of thing. And then we wanted to be able to say We wanted to be able to say things like, okay, if there's a specific app, we want to be to like filter by the app. And so those were like actually the main pillars that we built our index around. So so we do a few things. So it's like the the in the the the temporal relevancy was pretty easy to get from, you know, we just kind of do these scans of the database and try to con normalize any type of timestamp field that we found and convert it into something generic that we can match everything against. We would do the human readable text, we would like have all the information about where that data came from, like which app it came from, which database. So that's like useful context as well that we can like allow the model to like use in its search tools. But then there was also quite a bit that we did when we were sort of doing like destructuring the the text messages and and the raw text that your phone was like storing itself, which is We would do the named entity recognition. So for example, if if you if Google as a company is referenced or a speci a specific engineer at Google is referenced, you'd be able to like pull those out and then attribute those, like have a separate table of all these subjects and attribute those to different sources. And this became very powerful. It just basically is like just keyword search, but it became very helpful because a lot of times people ask questions about specific specific people in their lives or what did so and so ask me. But then also if the model is kind of like trying to answer a a multi-step question, it can it can do like a broad scan of of like who are who are your friends, like what do your friends talk about, then it pull then it finds like very common names like that like of of like your closest friends that you talk to the most. And then it can go deeper on those specific things and say like, okay, well what did Brett and I talk about three weeks ago? What did Brett and I talk about in relation to like the football game the other day? Like now they can start to become very specific and so it it was basically this case of we wanted to try to extract any type of data and build an index around our database or inside of our database that relates heavily to the types of questions that users ask and users ask a lot about time, they ask a lot about people and they ask a lot about like high level like themes. And so that's kind of where we had to pull the semantics, the people and the times all together. So you so the queries were like very composable this way.
Arjay McCandless: Where did you get that user data with no users initially?
Sam Hall: Well, we had a very like strong sort of culture where if Like everybody that that that and on our team was using an iPhone and I basically said, look, I'm not gonna make you use the phone, but if you don't wanna switch to the phone, figure out why and then make the phone better. And so a lot of it was just like we would use it a lot. Sometimes people would grab the phone from me and want to ask it questions. And so a lot of it started off as just like, what are the types of questions I ask? What are the types of things that like I as like a consumer at the things I care about outside of work, outside of like specific tasks that that are more like work related, like what are the normal things that I wanna get out of the phone? And then we did like then we started getting out phones to people once we felt we were comfortable enough with them and and then that even that was kind of like reinforced throughout the sort of experience of like our beta our beta rollout.
Arjay McCandless: Is your beta rollout still going? Can people like DM you and ask for a phone?
Sam Hall: If If you send me a good DM, like we talked about
Arjay McCandless: Okay.
Sam Hall: in the beginning, I'll send you a phone. You cause because the difficulty is that you have like we do not capture like nearly any logs on the phone because it is such a sensitive like piece of of your life, and we don't want to like risk ever capturing any type of personal information. So our like our improvements rely heavily on the feedback that testers give us, and they have to like kind of go out of the way to give that feedback. We can't just look at a log and say, like, this didn't work the way we. wanted it to and so and also we have to send you a $500 phone and so so it's kind of like if you want to test it I I do want to see a little bit of a little bit of passion about it. But we have a site up.
Arjay McCandless: Paragraphs plus video demo.
Sam Hall: Yeah, yeah. If if you send me a paragraph, I would I'd very likely send you a phone back.
Arjay McCandless: We'll see.
Sam Hall: Yeah.
Arjay McCandless: Okay, great. this has been an awesome conversation. Where can people find you, learn more about you, your company, all that?
Sam Hall: Yeah, I I mostly use Twitter. my username is Schmall S-H-M-A-L-L R
Arjay McCandless: Okay, I'll I'll link it.
Sam Hall: I link it, yeah. but but yeah, we have like our company Twitter, which doesn't have very many followers, so that'd be really cool to to b bump those numbers. and then our website's wafer.systems and by the time this comes out we should hopefully have a new page up, which I was actually supposed to do today. but yeah, we also have a new video coming out to Twitter soon, so hopefully that does well. but yeah.
Arjay McCandless: All right, awesome. Well, yeah, thanks for watching guys. If you're new here, drop a subscribe on the channel. And if you're watching on Spotify Apple, leave us review. Thank you, Sam. Maybe we'll have you on again sometime. Yep. Peace.
Sam Hall: Thanks, RJ. Yeah.