AI-Driven Auto Shops with Mastertech.ai and Data Observability with Monte Carlo | E2035

0:00Can I try a query while we're here? All right. Can we do a 2019 Subaru Outback? Why won't the passenger window entirely go up? Why does it get stuck halfway up and then make us want to scream? So there was actually like two TSBs that were related to issues with the power window for this model. So this is at least a good starting point for diagnosing this particular problem. I'm not going to lie. That's impressive. This Week in Startups is brought to you by Vanta. Compliance and security shouldn't be a deal breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast.

0:40Twist listeners can get$1 ,000 off for a limited time at vanta.com slash twist. Squarespace. Turn your idea into a new website. Go to squarespace.com slash twist for a free trial. When you're ready to launch, use offer code TWIST to save 10 % off your first purchase of a website or domain. And CalShea, the largest regulated predictions market now lets you trade on U.S. elections. Visit calshea.com slash TWIST to see live election odds, place a trade, and get$20 when you deposit$100. Hey, everybody. Welcome back to This Week in Startups. My name is Alex. I'm Alex over on X. You can also find me on LinkedIn pretty much anywhere around the web.

1:23But today I have good news. We have an excellent founder on the show, part of the Launch family. Here we have someone with deep technology experience that seems to have taken a bit of a turn away from her original work and applying startup magic to an entirely different industry. So please welcome to the program. It's Linda Gray. Linda, how are you? Hi, Alex. I'm doing well. Just really thrilled to be on the program. And yeah, thank you for all the support from launch. And it's been fantastic to help us, you know, kind of get traction and accelerate our company. Awesome. And I just realized I didn't actually say your company's name out loud.

1:59So let me fix that. Mastertech.ai. And Linda, I really wanted to talk to you, not only because I like what you're working on, but you spent 15 years at Microsoft as a principal software engineering lead, manager, other roles. You also worked at Niantic at the Pokemon Go game. And I used to cover Microsoft. I know that company. I have known many people there. I'm a little surprised that you went from those two jobs into what MasterTech is building. So first of all, tell us the founding story. And I can't wait to hear how this came to be. Yeah, absolutely. It's definitely not a straight line. If you look at my career from Microsoft to Niantic, and that was a little bit of a pivot in of itself.

2:41And then now to Master Tech AI, kind of building AI for mechanics, for technicians, you know, frontline workers in these shops. So, yeah, so my career, I pretty much joined Microsoft out of college. So, you know, kind of rose up to the ranks from intern when I first interned back in 2004. That was my first stint at Microsoft and joined full time, kind of worked my way up the chain to like I was principal engineer on the Outlook web team for a number of years, went into engineering leadership, engineering management. So I led teams at Microsoft Outlook, Microsoft Teams, Xbox Live and built a lot of great software, a lot of great teams and learned a lot through that process.

3:29I think for me, after 15 years at Microsoft and 17 years in tech in general, I really just didn't want to just do one thing in my life. And I was like, there's so much more that I'm excited to do. And I think being in a larger tech company like Microsoft and Niantic, it offers you a lot of great opportunities to make such large-scale impact in the products that you work on. you know the products that you know i'm used to serving like you know 200 million monthly active users right with kind of the office uh you know set of users xbox you know niantic etc but it is like hard to kind of really go do the true greenfield projects the zero to one like really build something new and innovative that you know big tech companies are not really going to necessarily want to take the risk to invest in.

4:27So that's, that's sort of what motivated me is like, I knew I wanted to eventually like really go do a startup, do a zero to one project. And, you know, and, and with kind of the innovation that was happening in AI in the last couple of years, the technology shift, it was like perfect time. It was really, you know, like, that's what excites me as a technologist, as something that we can really apply in the real world and make real world impact. Yeah. But Linda, so all that tracks. Yeah. Technology experience, lots of time, different teams, different projects, want to go out there, want to go zero to one, go into something greenfield.

5:08But why did you pick, you know, essentially auto shops, mechanics, and the care of cars as the place to apply AI? Yeah, absolutely. So, you know, so when I started looking into the space and looking at the potential problems that I wanted to solve with building a new product, building a new startup, you know, I was looking at sort of the gamut of, you know, the problems that I had experienced, you know, in my career, the problems I was familiar with, as well as problems like outside and other domains. And for me personally, I really didn't want to build the same set of or solve the same set of problems that, you know, I always see being solved in tech.

5:50It feels a lot like a tech bubble, you know, through all my experiences at Microsoft, and, you know, these larger, larger tech companies. And I was like, if I do a startup, I don't want to be like, you know, the 20th company that just solves another variant of this, this particular problem. You know, I don't want to give specific examples, but it's a lot of like tech companies and engineers solving problems for other tech companies and engineers. And I felt like there was this big rest of the world where there were so many industries in the real world that was underserved and overlooked, you know, and we can actually make such a bigger real world impact by bringing technology to these like underserved markets and actually have a bigger real world impact.

6:35So one thing, if I go back into my Microsoft reporting memory bank, and it's been a while since that was my core day job. But I recall one time talking to a Microsoft team, it may have actually been the team's team about how they were trying to bring the product out to more frontline workers. Yes. And if I think about frontline workers, people who are actually literally on the ground, working on cars as they come into shops and so forth, are about as frontline as you can get in the employment space. So is there any connection between the Microsoft frontline push and where you pick to build your startup?

7:07Yeah, definitely. And I was on the D2D platforms team on the apps ecosystem for teams. So, yeah, there was a lot of opportunities to integrate with third-party apps on the developer platform and, you know, to work with frontline workers, like, not directly, but indirectly through serving them. But it was still, you know, very approaching problems in a very horizontal manner. And that's sort of what you have to do in big tech companies. When you build like Outlook, when you build teams, it is sort of sometimes ends up being a least common denominator for what is kind of the platform that can enable sort of horizontal experiences.

7:47And I was more like, but it does end up being like a least common denominator experience for any particular set of problems. So I was more excited about like, hey, let's really just make the best freaking solution for this particular problem and use technology to its fullest. And that's what I was really excited about. Founders, do you want to sell to bigger customers? I know you do. You got to get that ACV trending up and you want to push your turn down, right? Sounds good. But to sell to those big buyers, you need to clear all of these compliance checks. You know that. That means you got to have things like SOC 2 sorted out.

8:25What's SOC 2? It's a standard and ensures that companies keep their customer data safe. And if you aren't SOC 2 compliant, you can kiss those big deals goodbye. You're not going to land the lighthouse customers. You're not going to be able to operate at the highest end of the market. But Vanta makes it really easy for you to get and renew your SOC 2 compliance. On average, Vanta customers are compliant in just two to four weeks. Can take months without Vanta. And they automate compliance for GDPR, HIPAA, and more. So you can sell to bigger customers in whatever markup your startup is going after.

8:57Vanta is going to save you hundreds of hours of work and up to 85 % on compliance costs. Stop slowing your sales team down and use Vanta. Get$1 ,000 off at vanta.com slash twist. That's vanta.com slash twist for$1 ,000 off your SOC 2. Okay. So I mean, like I've owned cars. I don't drive much anymore. My spouse is a great driver, so she does it, but we have to take care of our vehicles. We bring them into places, auto shops, dealerships, and so forth. I'm really curious how technologically savvy is the average car shop that works on vehicles? Because what you guys have built looks really cool. And I think we're going to do a demo here in a minute, but to me, I'm just curious how it translates to folks who are, you know, with a wrench in their hand and a socket in the other.

9:42So I get that question a lot. And I think that, you know, the, it is an industry that I think it's more with on the shop owners side of things that's more resistant to change. For the technicians, they're actually pretty tech savvy. And it's sort of in the name of the, of the job where to learn a lot of different tools, have to keep up with, you know, different things for different vehicles. And, and, you know, like a lot of them are Gen Z, They're like much, you know, they, you know, adopt technology. They are TikToking, right? They're like, so they're used to like, you know, all of these latest technologies.

10:16And if there just hasn't been a great tool and application that's been built for them. And that's what we were really excited about. So, you know, how I got into car repair, to answer your previous question, you know, I met my co-founder, Dave. He is a 20-year experience in auto repair industry, working as an ASE master technician, sort of the highest rank of being a technician and shop owner, so knows every pain point in the industry. So as we got together to kind of talk about what's possible with AI, it just really felt like the perfect application of AI. And we're really going to do it in a way that works on the ground in these shops.

10:54And we're seeing the adoption, which is really amazing. Like our engagement is going up week over week. It is being used on the ground in these shops by the technicians. So it is pretty cool to see. But before you explain what MasterTech does, I was going just through all your materials and I'm going to attempt a summary here. Essentially, if you are a technician working at a car that comes in, there are so many manufacturers, cars, models, different models, different years, different issues that can come up, different technologies. And so what if you had a place you could go and ask questions and the piece of software software using AI, I presume, could fetch the information you need and then present it to you.

11:31So that way, no matter what vehicle comes into your shop, you know what's wrong and then how to fix it. How is that? Yeah, I think that's a really good summary. So for the job, for these technicians and mechanics on the ground, the job is really two parts, right? It is one navigating all of this digital information and technical specification for the vehicles. And then second is the hands-on job that they're performing, like wrenching and replacing components and replacements, et cetera. So when we talk to the shop owners, they say that on average, 25 % of their technician's time is actually spent doing computer research because every single vehicle has a different set of specifications, procedures, issues.

12:18And it's a lot of just navigating through all of that data to be able to know what to do on the vehicle. And you have to be very precise if you put in the wrong amount of fluids for a specific engine that could cause a ton of damage. So it's kind of the analogy we like to make is MasterTech is sort of analogous to healthcare for people where there's a lot of AI investments in healthcare right now helping to coalesce patient history, diagnostics for doctors and you know, all of this data, like reading the test results, etc. So we're doing that for vehicles. But unlike humans, like, actually, each vehicle comes with a blueprint for, you know, all of its specifications.

13:08So, you know, like for you or I, it's like, we don't come with like, okay, here's all the parts, and here's all the numbers. And here's like, you know, what, you know, that's something unique for this human, right? Whereas this is all the information they have to work with on cars. And so it is about bringing that together. So in terms of the data that we are sort of... That's where I wanted to go with this because I was so curious where you guys got all the information because I presume it's everywhere in old books that are in glove compartments and old databases and manufacturing stuff that you can't get your hands on.

13:42So tell me about how you got all the data, not only ingested, but correct. Yeah. So the data is extremely important to get it right. So this is something that we were very conscious of from day one, because, you know, with AI, there's always concerns about trustworthiness, about accuracy. So we knew at the start that we did not actually want to train a custom model where, you know, this data is sort of baked into it, because at the end of the day, all like LLMs and ML models, it is still a probabilistic answer. And for a lot of these specifications and things, you never want to give an answer that's probably right.

14:23It has to be right or not right. So we made sure to get our data from the only real sources of licensed OEM data providers. So they license out the data on behalf of the OEMs, but we still have to get approval for our use case with each of the OEMs. And that's been an ongoing process we've been going through for the past year. But we have majority of the approvals that we need, where all of this data is directly from the OEMs for the procedures, specifications, and everything else. And we're using AI to navigate the user's intent and then pulling the correct data and then serving it to them in the way that can best assist them.

15:09So does the AI component of this allow a technician to essentially ask questions and then it parses that, turns it into a query, and then goes to the database of information that you have from OEMs and then grabs the right bits and then brings it up to them? Yeah, yeah. Something like that. We do have some, you know, kind of like just-in-time vectorization of the data. So it allows for semantic search and not just like sort of a keyword search for the OEM databases. So try to, you know, be as friendly as possible, you know, which these technicians are not used to. They're used to like very strict like file folder lookup for these documents or very strict keyword searches for full documents.

15:51So now it's really about having a conversation, finding answers, but every answer is backed up by the source and it's with the original manufacturer copy as well. So going back to the OEM data point, you said you have to get agreements with or permission from the OEMs themselves. And you said you had the majority of them. Do you guys need to have all of them or is there a certain like critical mass of, okay, cool. We have 80 % of the OEMs out there for cars. So now we have enough that we can take this out to, you know, the average auto shop and sell it to them? Yeah. So, you know, like it really depends on the shop, but, um, you know, we, we are able, um, to have enough value with all of the, uh, OEMs we have right now and really only missing, you know, two major ones with Honda and Toyota.

16:38Um, and those are big manufacturers. Um, I think honda might be might be pretty close um um that we're working with but um they're actually a smaller volume for our our primary customers which is the repair shops because they are sturdier cars so it's like it is a little bit of like okay you know it's it's actually like it's actually fine like it is a smaller volume for a lot of our shops and a lot of our shops are like they specialize in european vehicles or something like that i'm sure there's a bmw specific place if you have bmw information you're good to go but can i just say how hilarious it is that you see fewer honda and toyota cars because they work i mean that's really funny uh but those are both japanese car companies and subaru is as well has subaru come to terms with master built yeah okay so it's not japan problem it's not like those there's not like export rules from the japanese economy that disallowed this sort of thing no no so we just um we we have to get approvals from each oem separately.

17:39And aside from Honda and Toyota, we have basically everyone else that we need. And yeah, we're really hopeful that, you know, Honda and Toyota, like we're seeing some traction with our progress with the approval process, data compliance and et cetera, but it has been a process. But yeah, so that's like kind of the first set of data that we're focusing on is OEM data, which is really critical in these shops. Like I said, you know, it is a pretty critical part of the job to get all of this information before like doing the actual procedure on the vehicle. But the other two major sets of data we're focusing on is, you know, one, the community data.

18:16So, you know, like if you're a technician, right, like, and you're, you know, going from apprentice level to, you know, master technician level, it is really about like, yes, getting all of the OEM, you know, kind of procedures and specifications and knowing how to understand that correctly. But two, it is a lot of personal experience, right? Of like, hey, you know, I have, this is a person that's worked 20 years on Subarus and knows everything about them, all the ticks and all of the quirks and issues that's for the set of vehicles. So there's a huge component of the job where it is relying on that kind of human experience and community data.

18:53And right now, you know, there are some places where there's forums or, you know, some other specific places where they go and get this community data. But that's really going to be a focus of ours at the end of this year, launching a user submission content pipeline. And just the vision is really becoming like the stack overflow for automotive repair. I was just thinking that, you know, what would make your service not only very good, but also entirely unique and uncopyable would be to have the technicians that are using it leave notes, information, and breadcrumbs for people coming behind them because then you'd have the OEM data and the real world data, if you will, at the same time, which would be super powerful.

19:38Yeah, absolutely. And that is really the essential parts of the job right now, right? And we just want to build a platform that can, you know, so where you don't have to necessarily spend 30 years like working on one specific vehicle to gain kind of the insights and knowledge that that can be shared across to help everyone else that's working on these vehicles because this is a very hard job it is a very dangerous job as well so auto technicians actually one of the top 15 most dangerous jobs in america and a lot of it is because that you know there's so much pressure in production and not enough like software help or time to even find all of this precautions or information and safety and so it is about like you know really you know using AI but you know not really you know our our end product is not going to be artificial intelligence it is going to be sort of helping to accumulate human intelligence right that has all of these experiences that has been built up over the years from all of these technicians and, yeah, help to make the job faster, easier, safer for everyone.

20:50All right. Squarespace is the place to build beautiful websites. You know that. And you can bring your startup to the world with an amazing design and unbelievable functionality. And you don't need to hire developers or a web shop to do it. And if you do hire a web shop to do it, they're going to charge you an arm and a leg. It's going to be thousands to tens of thousands of dollars. They're going to disappear. They're going to change careers. They're going to ghost you. And then you're going to be left with a total mess and have to start over again. Well, you can trust Squarespace, who I've trusted for over a decade to maintain and really evolve and grow our web presence.

21:24Every business needs a gorgeous website, startups, schools, banks, projects, books, and you can now do it with AI. Squarespace is always at the cutting edge. They now have a design intelligence tool built into Squarespace. And so it'll get you a custom website built for you. That's gorgeous. Easy peasy lemon squeezy, as I'm prone to say. And we all know that your business needs more than just a great looking site. So they have built in payment technology. So if you build something, you can sell it. If you start racking up all those sales, you need analytics to figure out where it's all coming from.

21:58And of course, you want a great domain. So they got you covered there. Even better, you're going to save a little bit of money. And everybody likes to save a little cheddar and Squarespace is so generous. Squarespace.com slash twist for a free trial. When you're ready to launch, you go to squarespace.com slash twist and get 10 % off your first website or domain purchase. That's squarespace.com slash twist. We love them here. Longest running partner on this week in startups. And we really do appreciate that. So Linda, I would love to see this in action. It's great to talk about it, but it's a little hard for me to conceive of what it looks like kind of on the ground.

22:26So can you give us a quick tour? Yeah, absolutely. And as you do this, Linda, just for people who are on the audio version of this, can you just live sportscast as you go through and explain and detail kind of what you're doing? Yeah, sounds good. So, yeah, so as you guys can see here, so basically this is the MasterTech AI platform. You can add your vehicle into the platform that your shop is working on. It is fully mobile optimized. It's a web application, but we're going to be building a fully native mobile application in the future as well. So, you know, voice, vision, you know, the whole gamut for multimodal AI assistance.

23:09but once you have your and you can also scan in your vehicle when you're on mobile with a VIN scanner camera-based VIN scanner as well but let's say we already have this vehicle added in here for this 2010 Mercedes so it is a like a AI first chat first interface where you can get any kind of assistance that you need we do have some quick actions pre-built for some of the more most common actions for performing a procedure, diagnosing issue, looking at specifications, you know, managing labor times, you know, for your shop, etc. So, so let's just say we are diagnosing an issue today and we're, there's a noise that's on this, on this Mercedes that's coming into the shop and we're just going to say like, hey, let's do a noise diagnosis and get the kind of assistance that we need on that.

24:00So we're going to go through, we're going to pull up the the OEM procedure for noise diagnosis for this particular model. It actually comes with some detailed flow charts for what to do, which our AI can help you navigate as you try these different things. And it comes with some of the known issues for this model as far as noise issues that has It's been reported, you know, this is to NHTSA, National Highway Transport Safety Administration. And all the OEMs are required to, you know, report all of the known issues for any model, which we have fully indexed into our system as well. So it basically gives all of this information about how you can approach this generic issue and, you know, give some known issues.

24:50But, you know, we did give a pretty generic, you know, kind of ask. So, you know, the system detects is like, hey, we can actually narrow this down a little bit if you can give me a little bit more information. So when is this noise issue occurring? Is it, you know, under certain conditions? So we can say like, hey, it's happening. Let's say when the vehicle is accelerating, right? By the way, you're only putting in like noise issue and when accelerating, like you don't have to give it hyper detailed requests. It's pretty much just giving it a couple of words and then it goes to work for you. That's right.

25:26So when we add in the more specific information, then it's like, hey, based on the fact that it's happening when it's turning, here are some more likely causes of this issue based on all of the available documents. And here are some known issues that's like, hey, there's this one that's been reported that's known for this engine that there's this noise issue when you're going forward or reverse. and so it really kind of helps to narrow down kind of the specifics of what the technicians are looking for. So today, you know, without this platform, they're going off to all these different sources, database sources to try to find this information and cross-correlate and so it's really about doing it faster, doing it more accurately, more comprehensively and yeah, it's like, it's kind of what I think technology should be used for is, you know, in really helping productivity in the real world.

26:18so can i try a query yeah over here all right uh can we do a 2019 subaru outback okay so let's add that vehicle um 2019 so we'll add it by your make model this time instead of then um let's say this is the model it's the v6 version if that matters okay well hopefully this one is this is still good yeah because we actually do have a minor annoying issue with it okay so uh why won't the passenger window entirely go up why does it get stuck halfway up and then make us want to scream okay let's say why does passenger window get stuck perfect i i'm so curious if this is gonna work if this works linda i'm gonna jump for joy um so it looks like there were actually some known issues that were found.

27:15So it looks like the most likely issues is a faulty power window switch, a mechanical issue with the window regulator. So there was actually like two TSBs that were related to issues with the power window for this model, and as well as a Subaru procedure for how to like reset the module and work on it. So this is at least a good starting point for diagnosing this particular problem. So how has traction been in the market? And how much are you guys focused on growth today versus kind of still building out the product itself? Yeah. So we actually launched publicly to shops May 1st. So that was only about five months ago or so.

28:04so we've gotten so far about 40 shops signed up on monthly subscription yeah and we've actually the product has we've built a lot more data into the product a lot more features so you know with product market fit it's always an evolving process but it's been like really amazing actually to see the engagement from our users increase over time. So even the users, the same set of users are coming back to the platform as we add more data, as we add more features. And the engagement is growing week over week. That's per user, per shop, in addition to the new shops that are signing on. So it is really cool to see.