[00:00:28] Jay: Hi, everyone. Welcome to The First Customer podcast. My name's Jay Aigner. Today I'm lucky enough to be joined by Kelvin Pho. He is the CTO and co-founder of Mia Labs, the AI super employee built for automotive retail. By the way, that was different earlier, Kelvin, so I know you guys are testing messaging, 'cause I swear that was different earlier this morning when I went to check. Kelvin, how are you, buddy?

[00:00:47] Kelvin: Good, good. Doing really well. Great to meet you, Jay

[00:00:49] Jay: Am I right? Did that change or is, am I crazy? Didn't it say something else

[00:00:53] Kelvin: For-- No, we've always been in automotive. We've 

[00:00:55] Jay: No, no, but the slogan, like your hero on your website.

[00:00:58] Kelvin: Oh, yes, yes, yes. That did, that recently changed. Yeah, we upgraded it, changed the scope of it, and it's been great

[00:01:04] Jay: Love it. All right. So, where did you grow up, and did that have an impact on you being an entrepreneur?

[00:01:10] Kelvin: Absolutely. So I actually was born in California. My dad got a job at NASA as a rocket scientist for the space shuttle program, and he moved the family to Houston. So, grew up in Houston, went to high school there. And actually it did affect me a lot 'cause then my dad left the space industry and went into real estate.

So I saw him build his real estate business from the ground up in Houston. Did well there and then he... After I graduated to, and went to college at UT, he like, retired and moved to Florida, so. But yeah, it's been, that's kind of my entrepreneurship has always been in my family. So yeah, that's how I got started.

[00:01:46] Jay: What do you do differently that you know about than your dad did as an entrepreneur?

[00:01:53] Kelvin: What do I do differently? I think, you know, he was an immigrant from Vietnam, through the war, right? Vietnam War. And so I think I'm a lot more aggressive as an entrepreneur than he was. He, like, kind of stuck with, you know, things that he could touch and feel, like real estate assets that, you know, that were more cash-like, and I think I'm a much more aggressive entrepreneurwith higher hopes for what I build

[00:02:16] Jay: Okay, very cool. So, fast-forward to Mia Labs. What is it? Where did it come from? Why is it called Mia Labs is the first question, and then tell me what it's about.

[00:02:27] Kelvin: Yeah. So in 2023, when generative AI came out and to the masses, me and my co-founders, I've known my co-founders, mentored them through their first company at Techstars, got along really well, became really good friends, and actually going to one of my Airbnbs in Montana on a founder trip, we're like, "All right.

We know who we're gonna work with. We know who we're gonna build the next company with. What are we gonna build?" And so we didn't want to build something... We wanted something that had a moat, so we wanted to do something technically challenging, which is a voice basically generative AI voice agents.

And so we started prototyping in 2023, and what happened was, you know, we were like, "What should the name be?" It's gotta-- It-- We want it to be friendly, short, easy to spell. You can't mispronounce Mia. You can't, you know, you can't misspell Mia, and kind of utilitarian mode, like the domain was available, mia.inc was available, and so we just went for it.

That was kind of the genesis of the voice AI technology that we built everything on. And how we got into automotive dealerships was actually one of my co-founders started the AI club at UT Dallas, and a local dealership based in Dallas reached out to him and said, "Can you build a voice AI agent for my car dealership?"

And he says, "Actually, yes, we can." And that's kind of how the entire thing started, how Mia went into voice AI and into the car industry.

[00:03:52] Jay: And the rest was history. And I think we talked like randomly. We got connected like I guess a couple years ago, and then fast-forward and I'm like, "Oh man, look at you. Now you guys are like they're like a real company now, and you're all like..." It's incredible to watch just what I've seen from afar, so very cool.

I guess has the, has your thought on the moat changed now that technology and like com- you know, complex technical systems, you know, are available to, quote-unquote, "anyone"? Does that moat change for you guys, or has it changed?

[00:04:26] Kelvin: Yeah. So I'd say our moat started with voice AI technology, and I'd say in, to some extent it's still there. We have, you know-- We spend a lot of time. We have the best voice-to-voice latency in the industry, and continue to get better. As the models get better, the models get faster, we get better with our infrastructure.

So all of that we still have an advantage on, but, you know, pragmatically speaking, as a technologist, you know, that will, you know... The game is changing. The models are improving. They're becoming more commoditized over time. So we see that lead extending throughfor probably about a year, maybe more, for the core voice AI technology that we actually have.

Early on, we developed some techniques and patented them. They're processing at the USPTO as we speak. And so the-- But as far as the dealer, the moat in the dealership space, we've done a lot of automotive industry integrations, 40-plus different automotive integrations. These are hard to do.

You know, the documentation is, has something to be desired. And every integration is nuanced in its own way, and the thing about them is that, you know, since it's poorly documented, you've gotta go, you know, try the APIs and work around them in a lot of cases. So there's been a lot of work in the 40-plus different integrations, and that's its own moat as well.

[00:05:44] Jay: And has there been thought to that, like beyond a year when voice maybe becomes commoditized? Like is there a shareable plan around like how that you leverage all the stuff that you guys have built and the network you're talking about? Because obviously like that's everything, you know, relationships and like a physical network of people and human beings and dealerships and all that sort of stuff.

So I mean, whatever you can say, I would love to hear like how you plan on at least addressing that switch, 'cause it's gonna be a switch as in your core business is in some ways I would assume have to change.

[00:06:16] Kelvin: Yeah, absolutely. And so the key for us is the data that we're collecting right now. And so like the dataset of, you know, customers talking to-- people talking to people, all of that exists, but the dataset of people talking to AI agents over the phone, and specifically in our industry, that dataset doesn't exist.

So we have, today, we take about half a million AI conversations every month. And as that accumulates for our dealers, each dealer gets better. And then in aggregate also, the platform gets better for all of our dealers. So there-- we definitely have an advantage in speed velocity as far as the data moat that we have as well.

[00:06:53] Jay: Got it. How AI forward are you guys from auto-complete to harness engineering? Like, where are you guys on the spectrum, kind of just generally speaking across your engineering organization?

[00:07:08] Kelvin: Yeah. So my engineering org, is 30 engineers, so a combination of like, you know, your general software engineering and four deployed engineers. And so I would say like 90, 92% of our code is AI generated. Everything gets an AI code review, so we're on the bleeding edge of AI coding, I'd say

[00:07:28] Jay: Well then I'm gonna have to ask the natural question from a quality assurance guy. How the hell are you validating and how are you treating validation in this new world, as opposed to, you know, a couple years ago when maybe 10% of your code base was AI generated or none?

[00:07:48] Kelvin: Yeah, absolutely. So, there's a lot of work being done on like, you know, trying to get everything into the code base. So we have a monorepo. It makes it much easier to unit test as much of the functionality as possible, makes it better for the AI agent to write front-end and back-end code. The, you know, database configuration files and everything like that are all in the same repository.

and so it makes it easy for the agent, but also any documentation around how the system should work and, you know, the coding standards and coding patterns are in the repo as well in an MD file of some sort. Architecture diagrams, descriptions of like, you know, linting or coding patterns and things like that.

And so those are all in the repository, and that helps the AI code review agents like do a much better job doing initial review before we do like human review for all of our code reviews

[00:08:41] Jay: And last thing on this, because obviously I could talk about it forever, but how are you kind of measuring and/or understanding how many humans and what touch they need at this point? Like how, like just do you guys like, is it a standard kind of, hey, we run it through our, you know, people use it and they run a regression suite by hand and we kinda know it works?

Like what do you guys use humans for in that process right now?

[00:09:07] Kelvin: Yeah, we actually have humans, we have an engineer, review every PR. And so the code review process on that front hasn't changed. It has accelerated 'cause, you know, human reviewers are using AI to describe functionality and things like that on... As they review the code on looking for hotspots of major changes and major breaking changes.

But it's, it is everything does go through human review. So, my philosophy here is that, you know, like engineering has changed so drastically. But really my philosophy is like classic software engineering practices are more important than ever. While before you could, you know, wait for the code reviewer for something but it...

The documentation of how and describing how you want your system to behave and the quality that you need it is more important than ever. You know, testing is more important than ever. You know, good, you know, software design patterns are more important than ever because if you think about it, in the old days of software, it's like you want the ideal world in the SaaS era was you have everything documented in testing so you could hire an associate engineer and they could contribute to the code base as fast as possible and not break everything, right?

And so all those practices in place were to do that for, to be able to hire that talent. But now the, you know, the associate software engineer is a coding agent. And so, it's, you know, writing code faster, reading everything faster. So high quality code base, high quality patterns are more important than they ever were

[00:10:38] Jay: Yep. Yep. And I think the companies that I've heard be most successful agentically are the ones that essentially onboard them as human beings. Is they onboard them like junior devs, like you said. That's the... I think that's the right way to think about it, 'cause otherwise you just make assumptions and then, you know, quality gets kicked down the road.

And to your point, shift left, you know, all the other shit that, you know, test-driven development, that it was always a joke forever. Like now we can actually do a lot of that stuff, and it's like you can shift left because we're going so fast that it doesn't matter anymore. You can just actually move the quality up in the pipeline.

So totally agree and love that very much. All right, back to the business itself. So, was the first customer, was that the actual dealership? Did you guys make that... You guys made the voice agent for that first customer?

[00:11:19] Kelvin: Yep. Vanguard Kia of Arlington. We hold it dearly in our hearts.

[00:11:24] Jay: I love that 

[00:11:24] Kelvin: So ourScott Traylor, our dealership, basically, the dealership operator who found us, was our fourth partner and he lives in Dallas and knows the Dallas network really well. So we started with them, incorporated the company within two months, launched our production, taking our first AI phone calls, and in five months signed our first paying contract with the Vanguard Auto Group.

So that was, that was the start of the story there.

[00:11:48] Jay: I love that. And so who's your customer today? Is it the 

same persona, same type of same, you know, if you're going for new business today, is it the same one as, as you were back when you started?

[00:11:57] Kelvin: Yeah. Automotive industry, they break it up into three segments. We do. We have the majors, the strategics, and the franchise. And so basically, you know, larger enterprises, they operate differently from the mid-sized automotive groups versus, like, the auto groups with, like, 10 or fewer, right?

And so each one of them are different. We have different processes for each one to basically serve the dealer in the best way possible. And, but yeah, we've got, you know, 500-plus dealers, signed up with Mia. We've scaled phenomenally over the last two and a half years. And yes, yeah, super excited to be able to serve the dealership space.

And there's-- Running dealerships is a really, really difficult thing. You know, it is a very hard business, and our whole thesis is to make running a dealership as easy as possible and automating as much of the grunt work as possible for the dealer

[00:12:47] Jay: Love that. All right, well, let's talk a little bit more aboutyou. Do you care about personal brand at all? Do you get out there and, like, try to do... I mean, obviously you're on this, so that's a yes, a little bit. But then generally speaking,how much do you get out there? Are you 100% Mia Labs when you're out and about, or is there a Kelvin that's, like, not Mia Labs as well?

[00:13:09] Kelvin: There is a Kelvin that's not Mia Labs, but it does-- has nothing to do with tech. So when it comes to work, all of my, all of, like pretty much everything I do for Mia Labs. So I don't really... Honestly, I don't do a lot with my personal-

[00:13:22] Jay: Okay. Well, hopefully this is the start. We can kick you off. We'll get you all sorts of clips and stuff, and we'll get you out there personal branding Kel- Kel, what do you like to do? I mean, I'm, this isn't even part of the show, but what do you like to do outside of Mia Labs? What does Kelvin do for fun?

[00:13:35] Kelvin: Yeah, so I actually, I've been doing this for nearly 20 years. I'm a international competitive ballroom dancer. So I just came back from the UK, and my partner and I took seventh place out of 125 couples worldwide. So yeah, basically if I'm not working, I'm at the dance studio practicing and training, basically like an amateur competitor, and that's kind of what keeps me balanced in life.

Like physical activity, it helps a lot with my stress levels and balancing everything and it needs to be an activity where I can't look at my phone. I can't, you know... Nothing's gonna distract me. I can't get notifications, so it has to be something that I put my phone down, so

[00:14:12] Jay: I love that. I think you told me that, a long time ago, and I think I said that was what the... Yeah, we talked about this. The Tim Ferriss book, "The 4-Hour Workweek," that's how that book starts, is he was winning a, you know, international salsa dance or some sort of dance competition with his wife or whatever, and I was like, "That, that sounds about right for Kelvin." All right. So, I have one more question about business. Are you planning-- Did you guys build this to sell it? Like, when

you sat down, when you sat down together to build this, was there a plan to exit at some point?

[00:14:43] Kelvin: Yeah. Like, so I would say like, like at some... Like everycompany wants to exit at some point, whether you go down an IPO path or you sell to a larger company. I think both are, both are attractive. I think what we need to focus the most on in this phase of the company is just add as much value to the customers as possible and build the best possible product.

And I think those, those opportunities will comein a few years' time, for when we've created enough value where we feel like we're at a point where we can exit at some point. So, I wouldn't say that we started the company with a specific like exit strategy in mind

[00:15:18] Jay: Mm-hmm. Well, so what does your day-to-day look like as CTO? I mean, the world is literally upside down and like Kelvin, one of my other favorite people from 7CTOs,he was like, "What you do is the only thing that keeps me up at night," when he's talking about, you know, the quality and quality assurance.

Like, what is your, what does your day-to-day look like? Is it just like, oh my God, here's the next new crazy tool we can use or update or problem? Like, what is the day-to-day life for you right now running the company and, you know, running the tech team?

[00:15:48] Kelvin: Yep. And so I'd say I think one of the key transitions and differentiators we have is we're really bought into like an engineering first operational model. And so basically when a dealer signs up, we assign an engineer day one. You know, AI is hard, you know, it's very complicated. You know, we as technologists are like submerged and surrounded by AI all the time, but our customers aren't necessarily that.

So we put an engineer upfront to make sure we're training the dealer on AI, we're explaining things to them in the right way. We build the Mia agent specifically for every single dealer, and customized for that dealer to make sure that we're solving the dealership's immediate problems that they're looking for to change their business whether it is appointment bookings, you know, containment rates, you know, inbound sales or anything like that.

So, we're doing,we're... That motion is,it's difficult to do and it's difficult to do efficiently. And so I spend a lot of my time focusing on that motion of like engineering led ops. So I think that's one, one core part of what I do. The other core part of what I do is, like you mentioned, like staying ahead of AI and making my entire engineering organization efficient.

Tooling the-- There's new engineering tools coming out every day, new models coming every day, so keeping on top of that is really critical for any CTO in this day and age. And I think the third thing that I work on, on a day-to-day basis is like, major partnerships. So we've got a partnership team and representing the company's technology and AI strategy from a partnership standpoint for us to like expand our reach and do better by our clients is really important.

So I think those are the three big things that I focus on.

[00:17:28] Jay: How do you even manage one of those? Like the middle one, the AI piece seems insane. How do you stay-- Well, first of all, I don't know if anybody's staying ahead of AI, but, staying up to date on whatever, you know, like how are you even doing that? Like, do you just, what do you just like refresh Twitter all day?

Like, I don't under- how do you even know what-- Where do you even look to know what's the real stuff to be paying attention to?

[00:17:50] Kelvin: Yeah. So like a, a lot of our engineering team and our culture is like really AI forward. A lot-- Like many of our engineers, most of our engineers are watching this stuff on a regular basis, and we just cross-bottom link the information that we find. Also like AI summaries on a daily AI summaries of the news for AI posted into like your personal email inbox or something like that.

Super, super, super, super helpful. But really like just building the culture that where it's everyone's job to stay ahead of AI is really important 'cause I-- it is very hard for one person to stay ahead. And then also, you know, trying everything is really important too. Music's creating the bandwidth to actually experiment with these tools is really important to find out what works for you.

As you know, you can read Reddit and you can read Twitter and what works for another organization with a different size, with a different industry, with a different operating or, you know, engineering motion or operations motion, it's-- everything I think is really pretty bespoke right now.

So you've really gotta try what works for your culture and engineering team.

[00:18:51] Jay: Can you even put together like a technical roadmap right now though? Like how do you even adopt something new right now with all the stuff go- Like if, you know, if there's a new tool and somebody comes to you, like is there... Like what's the evaluation process for you specifically as the head of engineering?

Like you're the one that has to kind of finally bless it. Like how do you get enough information about whatever that is and the context that it needs to be in to go, "Yes, that's the right tool for us to go spend money on"?

[00:19:18] Kelvin: Yeah. And so like first like, like I would, I think me and the team will try anything. And so like just putting aside, setting aside time for like one engineer to go prototype something or just try it and come back to the, come back to the team with the ability to do it, I think is important. But also creating a network of other CTOs, other engineers, other startups, who are trying these things all the time to figure out and kind of helping filter out like signal from noise.

Because generally speaking, like the experimentation you can do as a person or as even one team in one company is not enough to keep up nowadays. So I've got a deep network of other startups that are doing kind of in a similar mindset that I lean on.

[00:19:59] Jay: That's a great answer. I love that answer. I think that's like, like you said, maybe the only answer right now.

[00:20:07] Kelvin: Yeah

[00:20:07] Jay: What do you mean you can't rely on like the main, you know, whatever, I'm not gonna say mainstream media, that term means a lot of different things. But just what, like you said, you can't just like look on the internet and like get the answer, and you can't look at your small team and get the answer.

So yeah, there is like a, it's like a filtered community does feel like where... And I mean, I guess like in history, when you look back at a lot of those big nerd forums and tech movements happened 'cause these guys in some small, you know, kind of got together in that group, like you're saying, and shared some stuff about new technology.

So may- that's a great, I love that answer, and it's why I love Sevens for what the, you know, Sevens is like that kind of community and group. So, that makes a lot of sense. All right, I have one more question. This is not about business. I like to say, "Don't tell me you'd be doing what you'd be doing already."

But if you could do anything on earth and you knew you wouldn't fail, what would it be?

[00:21:01] Kelvin: If you could do anything on Earth and you wouldn't fail. It's like a work-related question, like a 

[00:21:05] Jay: Non-work relat- non-work related question. Just Ke- nothing business, nothing work. Just Kelvin, if you could walk out the door right now and do anything on earth and knew you wouldn't fail, what would it be?

[00:21:18] Kelvin: I think my answer is sleep, actually.

[00:21:21] Jay: Dude, that's a, that's the only time I've ever gotten that answer, and it's the best answer ever. I have

six kids and 

I need to sleep. I hear you 

[00:21:28] Kelvin: Subscribe to sleep as the number one performance-enhancing drug. That, the concept of that is, like, nothing else you can do, supplements you can take, you know, anything like that, or daily routine. I think sleep is, like, the number one thing you can do. It's where everything resets for the day and you can start fresh anew.

I think one of the hardest things for running a business and having a lot of competing demands is you have to make hard decisions that have a big impact really quickly, and if, like... And a handful of those probably happen per day for me. And so, like, being rested is the most important thing, so I just really, like, at the end of the evening, like, I think about the next day and I'm like, "Am I doing the right thing right now so that I can make the best decisions tomorrow?"

is how I frame it. And a lot of, you know, scrolling is, doom scrolling is not the right answer. Like, putting phones down is really important for getting ready for bed. So, like, I frame my evenings around am I setting myself up for success tomorrow? 'Cause who knows, at a startup, who knows what decisions I have to make tomorrow that could changeand it really deeply impact the company for months or years

[00:22:35] Jay: What a mature mindset, and maybe the coolest answer I've ever gotten, 'cause nobody ever said sleep. It's such a great an- I lo- I... God, it's such a great answer. That's great. Kelvin, you were fantastic. You didn't need your cheat sheet. You knew all about yourself. You did great. We didn't need your, we didn't need your handler.

He hasn't said a word. He's been quiet, and you did great. You did awesome. If people wanna hear more about you or talk to you or reach out to you, how do they reach Kelvin directly?

[00:23:02] Kelvin: You can-- I'm on LinkedIn. You can reach out via LinkedIn. And, yeah, I think that's probably the best way

[00:23:07] Jay: That's fine. You can give them one shot. If they don't, if he doesn't answer you on LinkedIn, your stupid spam message was not good enough to connect with. So try again, what is it, like three weeks later or something. All right, Kelvin, you're fantastic, dude. We'll link everything in the bio.

Like I, I legitimately, am rooting for you guys, and like I'm enjoying watching this, and like it's fun, and I like, I happened to stumble into like watching this really cool startup story play out, and I will be continuing to follow it. And thank you for giving us some of your time, and have a good rest of your week, buddy.

We'll talk to you soon, all right? Thanks, Kelvin.

[00:23:41] Kelvin: Thank you.

[00:23:42] Jay: See you, buddy. Later.

[00:23:43] Kelvin: Bye 

​