Voice AI Becomes Personal Infrastructure

Machine-generated transcript from the episode video, cleaned for punctuation and names. Speaker labels are not included.

[0:05] Welcome to the GTM Vault Podcast. I'm your host, Rick Koleta, and today we're joined by Dawson Chen, co-founder and CEO of Martin, a Y Combinator-backed startup revolutionizing the personal voice AI space. Dawson's journey from working with NASA and Stanford to building his own companies is nothing short of inspiring. In this episode, we'll dive into his experiences, the development of Martin, and the future of personal voice AI. Dawson, thanks so much for being here. Thanks for having me, Rick. So I'll get right to it. What inspired you to co-found Martin, and how did you identify the need for a personal voice AI in today's market? I've been a power user of Siri and Alexa for a long time, and I've just suffered at watching them be terrible for so many years. And the reason I'm such a power user is because I think my threshold for adopting these products is really low,

[1:07] and I really like voice as an interface. So with the advent of LLMs, we thought this would be a great chance to maybe push the state-of-the-art forward a notch or two. And it just so happened that the current products are all taken by these really slow-moving big companies, and we thought that with LLMs it would be really slow to turn the battleship around, and it'd be the perfect timing for a startup to jump in. So we decided to build Martin. And what have been some of the biggest challenges you faced while developing Martin? It's definitely a technical challenge to integrate with so many tools and then also to have a memory of the user. So with each integration, we try to make it really deep, and we want to basically allow the agent to use the software, such as Google Calendar or Gmail, use it as well as a personal assistant would be able to. So for Google Calendar, that looks like being able to not just read and write events, but also change a bunch of small

[2:12] settings. So privacy settings, or secondary calendars, or availability slots, or adding a Google Meet link. So having the agent being able to manage all these small parameters and be able to anticipate some of the user's defaults, or whenever the user specifies something specific, they can be able to tune it. For each integration, doing this is definitely a challenging thing. LLMs are not as good as you might think at filling in a lot of these small parameters, so we had to do a lot of tuning for that. And then for getting to know the user and building a personal relationship with each user is also a huge technical challenge. So it's not as straightforward as just answering a quiz about the user. You sort of have to anticipate what they want you to do based on their past routine, or the things that they're anxious about based on your conversations together, and the past actions. So we try to make those two layers really well, the integration

[3:13] layer and the memory layer. Got it. Yeah, that's a lot. And how do you see the space, the personal voice AI, evolving over the next five to ten years? I think now still most people struggle to onboard into AI products, and they're just not used to using natural language as an interface. So the AI tools are still only adopted by a small portion of the proportion of the population. But I think over the next five to ten years, everyone will start using AI tools, and the interface will become a lot more intuitive. So we want to ride that wave and be the personal AI that's the most intimate and has the most personal relationship with the user. And what that looks like is an agent like a butler that understands your daily routines, people who you interact with every day, the tasks that you're currently most worried about, being able to manage those and juggle those, both for scheduling purposes and also for communication, with your colleagues or your co-workers. And then we also want to take control

[4:16] of the most frequently used software in your life. So if you're someone who lives on Slack, all your notifications coming through Slack, we want to be able to monitor those for you, and if someone, a high priority alert, comes on, we can text you and then text your co-workers about it, or call you over the phone and explain it to you. So we want to be the closest to the end user, and that's the agent that we want to become. What was your experience like at Y Combinator, and how has it shaped Martin's trajectory? YC was basically the start of our company. We actually found our idea during YC, so without YC we might not have even started. Being in YC really taught us that we could aim really big and not shy away from super ambitious ideas. And so we picked what we thought was the most exciting, what we're most passionate about, and just go all in for that. What differentiates Martin from other voice AI technologies currently on the market? So we have a couple important layers. One is the

[5:19] interface, and one is the integrations layer, and then the third one is sort of memory and personalization. Interfaces, we allow users to call Martin over the phone, text him, WhatsApp him, email him, or speak to him in our voice app. So we want to make it as easy as possible to reach your agent. And it's also not a trivial task, because you want to make sure the phone calls are good quality, and make sure the texts, you should be able to text them from your phone, from your laptop, and group chats. So we try to make the interface as intuitive as possible, and you can also forward an email to Martin, and then soon you'll be able to add him in Slack or add him on a Google Doc. So that's the interface layer. And then I think another big differentiator is we have really deep integrations. So if you use Siri, you can't use Siri to add a Zoom link or add a Google Meet link to a calendar event, and you can't even sync Siri with anything outside the Apple ecosystem. So

[6:26] you can't use Google Calendar, you can't use Gmail. And we do integrations across all platforms, and each one is super deep. And then the last one, for memory, there aren't really any AIs right now that understand a user well to the degree that they can predict what you're worried about today and then try to surface some tasks they can do for you. And all these three layers play together. You can't just have one or two of the layers, you have to have them all together to be a really successful agent. So I'd say these are three of our biggest differentiators. You previously co-founded Campus Guide. How did your experience with that venture influence your approach to building Martin? Campus Guide was a small side project during college. We called it, it was like Airbnb for college tours. So you would match with a student at a university that you wanted to get a tour at, you would pay them through our platform, and you meet them on the day of your tour. It was a really good experience for me to learn how to build an MVP and also how to get users. So we hacked together an MVP really quickly. It

[7:29] was probably less than a week, and then we got our users through all sorts of guerrilla tactics. So we would stand by the admissions office at universities, and if you forgot to book a tour or if you missed the last tour of the day, we would recruit you and try to give you a tour. Great. We learned a bunch of do-things-that-don't-scale tactics from it, and I think it gave me the confidence to start something more ambitious like Martin. What are some of the key lessons you learned at Campus Guide that you've applied to Martin? Well, the core skills I learned were how to build an MVP and how to get early users. I think it also cemented in my brain that you should do things that don't scale. So you shouldn't try anything fancy in the early days, like paid acquisition or anything scalable. You should just try to get users one by one

[8:31] with a lot of Campus Guide. That's great. And with your technical background in blockchain, smart contracts, and computer vision, how have these skills influenced the development of Martin? I guess I just always hacked on whatever the new technology was growing up. So when I was in middle school I did a lot of iOS stuff, and then in high school I did some computer vision, and then in college I did an internship at a crypto company. So I've always been interested in just hacking on whatever the new technology is. So I guess the main thing that taught me was just how to pick up a new, sort of, when there's a platform shift, how to pick up on the new technology really quickly. So I feel like I've done that every time I've had the chance. How did your work as a citizen scientist at NASA and your research internship at Stanford contribute to your technical and leadership skills? Yeah, those are both internships I did in high school.

[9:33] Yeah, they definitely taught me to be resourceful and how to work on my own, because in both of those projects I was given very little help. I was sort of on my own. So yeah, and taught me how to just learn by Googling, and learn by emailing people, and reading papers online. Yeah, I wouldn't say I apply any of the skills, any of the core technical skills, to be honest. I feel for most startups, your technical skills from past jobs, besides the basic web dev or iOS app development, more than that, I feel like it hardly applies, especially if you're working on something consumer. Got it. What role has your education at Yale played in shaping your entrepreneurial mindset and technical expertise? I spent one year at Yale and then I dropped out after freshman year. It was a very transformative year, not really from a technical perspective, but more just I had a lot of fun and I made a lot of friends. I think if I went straight out of high school and did my startup, it

[10:34] wouldn't have really worked. I sort of had to meet a lot of new people, and I feel all my closest friends were really different from me, so none of them were really engineers. I sort of figured out what my values were, why I wanted to do a startup, how I wanted to live my 20s. I think I sort of figured that out in that first year. And I don't think I learned much, I don't think I got much out of the academics, to be honest. Maybe I should have gotten more. How did your involvement in extracurricular activities, like being student council president at the Harker School, help you in your entrepreneurial journey? That one actually helped a lot. I think being on student council helped more than any other thing I've ever done. I think our school had a much more involved and hands-on student council than most other schools. So we would host fundraisers, or events, or we'd do talent shows, or

[11:38] whatever. We would be doing something almost every week, and it was a lot of work. And the student council, when I was a senior, it was like 30 people, so you had to manage a lot of people, and you figure out what everyone's priorities should be and then point them in the same direction. And that taught me a lot. And it was also a lot of hands-on work. It wasn't just empty management or giving speeches. It was, we had to hit this, we had to fundraise 20K for this Ukraine fundraiser we were doing, and it had to be done. We have to find a way somehow, so we tried everything we could. So student council helped a lot, and I did student council for probably six years in total. I started in middle school, so it was definitely the most influential thing, I think, professionally. Sounds like the sooner we can take on responsibility and leadership roles, even if that be in middle school or high school, the better, huh? I guess so. Yeah, I think that's true. What are your

[12:44] long-term goals for Martin, and how do you plan to scale the company in the coming years? So our long-term goal, our vision, is to build an AI like Jarvis, basically. We want this agent to be extremely personalized and have a deep personal relationship with you. So if you just got assigned this big new project, the first person you would talk to is Martin. You would be like, hey Martin, let's go over this. I have so many things I have to do. Help me plan this out, put all these things on my to-do list, put these on my calendar. And then over time, through the next three or four weeks of your project, Martin will keep you on track the entire time, and he'll not just send you calendar reminders or whatever, he'll actually proactively be like, hey, I found some new research papers on this topic that we were talking about the other day, and here are some summaries, here are the links. Just do things proactively. And I think our long-term vision is for this to be open for every consumer. I think in the early days, our earliest adopters are these AI prosumers, and I guess you call

[13:48] them hobbyists. They have a really low threshold for learning how to use these AI tools. But I think over time everyone will have to use them, and it should be so intuitive and so magical when you use it that just after one week of using Martin, you'll be like five times more productive. And it's the same experience that you have after taking a Waymo for the first time. You'll just be blown away. You'll feel like you're living in the future. That's the experience we want to provide, and we want to be the most personal. What advice would you give to aspiring young entrepreneurs who are looking to start their own tech ventures? Yeah, I feel like a lot of this advice has been said many times. I think you just need to build confidence yourself somehow. That's the key, right? If you have the courage to start, that's all it really takes. So you do whatever it takes to build that confidence in the early days. For some people, that looks like doing some side projects. For me, I got most of my confidence from running Campus Guide and running student council, I think. And then I would say you

[14:51] should pick your sort of mentors, or your buckets of knowledge, wisely. There's a lot of misleading or just bad marketing now on the internet, sort of misguided startup advice. I think you should take your advice from other quality sources, people who have founded companies themselves. Awesome. Thanks so much for joining the pod. It was a pleasure having you on. Yeah, thanks so much, Rick.