Skip to content
Gun.io
August 1, 2023 · 45 min read

Season 4, Ep. 21 – Founder to Founder: with Massi Genta, CEO & Founder, Metabob

This week on Founder to Founder, Teja sits down with Massi Genta, CEO and Founder of Metabob. They talk about the process of building companies. the difference between using LLMs and GNNs to help with code reviews, and why adding context and a human element to your tech can help build a successful business.

https://metabob.com/

https://github.com/MetabobProject/metabob-vscode

Listen:

Read transcript

(THE FRONTIER THEME PLAYS)

Teja (00:00:04):

Yo, what’s up y’all? Today we’re chatting with Massi of Metabob. They use graph neural networks to conduct contextual code reviews. We get into sort of the difference between GNNs and LLMs. We talked about how Massi kind of started in entrepreneurship, found his way, basically to the cutting edge of using AI to solve business problems. Super interesting conversation; learned a lot. This is like, this is the second interview that I did on the same day with like, another genius. So, <laugh> I think you’ll dig this one, super interesting one. So appreciate y’all and let us know. Bye. (THE FRONTIER THEME ENDS)

Teja (00:00:54):

Tell me a little bit about yourself. Like, where are you from? How did you get interested in, I mean, programming, entrepreneurship? Just gimme some background on you.

Massi (00:01:03):

I’m originally from Italy, Northern Italy, a town right next to Torino. I don’t know if you know, but it’s in the north side, pretty close to Milan. Initially, I got interested in programming when I was in my teens. Specifically, I started to become very interested in the open source side of it, (Teja: Mm-hmm <affirmative>.) so started contributing in different projects. It kind of ranges, but yeah, I definitely became an integral part of the open source community, both initially in Italy, then in Europe. Like, I spent some time in London, and that was like, really what got me into programming in general. Just became passionate by myself, kind of self-taught in the beginning. So mainly out of YouTube and just again, meetups, and all of that. And then like, what got me into entrepreneurship was really like, I wasn’t chasing entrepreneurship, per se.

Massi (00:02:09):

It’s more like, you know, as a consequence of becoming passionate about programming and started to just build stuff for fun, and most of that was pretty much pointless from a commercial standpoint. (Teja: Yeah <laugh>.) But until like, one became like, my first company. It was with a professor and a friend in Italy. And it was, again, started off on projects I developed for school and classic story, I guess. I got asked to make it, to actually put it like, make it a company, gain traction, raise a bit of money, and that’s really how I started, but unlike a lot of entrepreneurs like, I didn’t really start with the classic like, “Oh, let’s try to find, identify a problem, a solution to the problem,” business plan. I just build something that had some success in the early stage, and someone with way more experience kind of mentored me to bring it to market, right? (Teja: Of course.)

Massi (00:03:19):

I made a lot of mistakes, though, in <laugh> all my initial projects that I wish I now know. But yeah, that’s really how we started, and we raised money in the United States in this area. That’s why I moved to the Bay Area, to also, one of my ventures, then I study here as well. After that experience, I was like, maybe I need to get some experience in business as well. So, went to business school, well, off the record really, like, it’s not really, doesn’t really teach you anything about business, business school per se, but <laugh>, it’s just like, a good way to enjoy a couple years of your life, I guess.

Teja (00:04:01):

Yeah. I feel like business school teaches you like, how to be friends with rich people. I feel like it’s good at (Massi: Exactly.) teaching. Yeah, that’s it, right <laugh>?

Massi (00:04:08):

Yeah. Especially like, I went to school here in Menlo Park and so, yeah, it’s exactly as you say. It’s really like, it teaches you. I came from a very like, mid-class Italian family with like, you know, I had to work pretty much my entire life, and all of a sudden, I was surrounded by billionaires, and I was like, “That’s a different lifestyle, I guess.”

Teja (00:04:34):

Yeah, no like, you go to Vail on the breaks, you kind of have friends that go into all these spots, and I say, “This is what happens when you’re…”

Massi (00:04:42):

Yeah, it’s a bit of anxiety, right? Like, everywhere you go, you’re like, always stand like, you know, dinner is not gonna cost me like, two months rent tonight. But, yeah.

Teja (00:04:56):

Totally. Yeah. So what open source projects did you work on when you were first getting started?

Massi (00:05:03):

I mean, a lot of kinds of <unintelligible> I’ve done. So, and then like, I joined a group. We started a project, actually open source, called Clyste, where that was really like, one of my first bigger projects with actually our current CTO, and the goal there was like, building a framework and a governance system to enable actually open source projects to run better and actually to enable monetizations to open source projects. So this was kind of our goal, because I’ve made a lot of friends all around the world, right, that were contributing a lot to projects, but obviously it’s hard to make money in most cases unless you get sponsored. And so that was kind of our goal and to kind of build a framework where first you select, we come up with like a, we can call it a “governance system”. Like, almost like a constitution, we call it, where, who is making decisions to each commit and how the projects needs to run.

Massi (00:06:11):

And then following that, really like, based on the value of the contributions and how much the contribution is used over time, we build a model to predict how much each contribution is worth from a monetary standpoint, and we kind of work with a few bigger companies that implemented it to actually pay their contributors. So I actually went “Okay, I think it’s still, on a certain level, running. Like, people are still sometimes using the constitutions.” But yeah, that was kind of like, the, for sure, like, one of the biggest projects I led. I joined also a lot of groups. So there is this group called <unintelligible> that I’m still a community leader. It’s based in Europe. I mean, now they do a lot of things here in Mountain View, but it’s for, you know, Python projects and a variety of like, events that we organize. Really like, trying to announce the open source side of it.

Massi (00:07:18):

And yeah. Recently, we actually worked with few groups in India as well. This specific group called WeMakeDevs, which connected to also what I’m doing right now at Metabob. It’s a very big part of our business, working and partnering up with open source communities, contributors, influencers, just to make sure that, you know, they give us feedback, and again, I’m very committed to it, also in the hiring process. And so we partner up with the group, we do many hackathons, so we sponsor, usually hackathons, as well as we try to do like, a few events per year where it’s not about the company, per se, it’s just about trends and like, hiring tips and all that.

Teja (00:08:13):

What’s funny is like, just earlier today, we had a conversation with the original founder and maintainer of Homebrew. He’s now got a, yeah, super cool guy, and he’s now working on this project called t.xyz, and it’s pretty cool. They’re also, I mean, they’re trying to also solve like, this problem of like, underfunded open source projects. It’s a thing, and it’s funny, we had a conversation. Like, there’s a sentiment in the open source community where like, they actually find money like, kind of grotesque, you know? Like, a lot of folks just, they’re like, “No, I don’t want any money. I wanna do this, because I love it,” you know? So it’s just, it’s interesting, and that probably gets in the way sometimes.

Massi (00:09:03):

Yeah. But you know, for big framework like, a lot of the things I was doing, it was like, also in the beginning, especially self intensive flow, like, you know, the big Python projects (Teja: Right.) like, jungles, all of those. Like, I think that’s really where I see the bigger opportunity when it comes to like, (Teja: Right.) our user contributors to monetizing, but anyway….

Teja (00:09:26):

So did you move to the Bay Area for Metabob, or were you already here, and then founded Metabob here in California?

Massi (00:09:36):

I moved here for one of my previous ventures. For Metabob like, I was already there, so I moved in the Bay Area in 2012 or 13. So it’s been already about 10 years, and Metabob, we started in 2019. (Teja: Gotcha.) So it’s been like, I was already there at the time. It was right before Covid.

Teja (00:09:59):

And then also how did you come up with the name? So, you know like, in physics, “bob” is like, a thing. Is that the origin of the name? Or like, how did you…

Massi (00:10:06):

It’s a mix of that and “meta”. When we first started, first of all, Facebook wasn’t called Meta when we started, (Teja: <Laugh>.) and then “meta” was like, you know, there was like, this concept of “meta programming”, (Teja: Yes.) which is really like, out co-generation, right? It was kind of a combination of the two that we put together. It was kind of random, but I liked how it sounded, and I was like, “Yeah.” Plus, “bob” like, I don’t dunno. I feel like we have a couple of team members that represent “Bob”. I dunno; it’s a cool name. So we were like, “Yeah, let’s do that.” So it was kind of random. Same for the logo. We literally drew it like, at the time I was living with my CTO and the UX designer. We were just like, imagining how Bob would look like, and that was kind of the look of it. Like, I drew it on a piece of paper and our designer then made it a logo.

Teja (00:11:05):

I’m looking at it. It does look like your classic like, kind of wisened neck-beard programmer, you know? When you imagine like, the fucking genius who knows where everything’s buried, that’s it <laugh>.

Massi (00:11:17):

Yeah, exactly. Yeah, we just need to, right now we are actually like, thinking about like, as we’re getting more and more enterprise customers, like, and growing, like, we start hearing, you know, that being like, a male, it’s kind of screaming at part of the community. (Teja: <Laugh>.) So, you know, all of that. So that’s what happens, right? As soon as you become like, from like, a hacky projects to actually like, gain like, visibility, then people start calling you out on everything.

Teja (00:11:52):

It’s a thing in a modern day company like this.

Massi (00:11:56):

We might, it’s more like, the logo might become like, a cat. Maybe not a cat like, but like, some type of animal maybe. I don’t know. We’re thinking about it, what’s best, but yeah. We’ll see.

Teja (00:12:08):

Yeah, it’s tricky, because it’s like, you know, you wanna be more inclusive, but you don’t wanna do it in a way that seems to alienate like, the original ethos of the company. And it’s like, it’s always, it’s like, it’s cool. It’s the company’s identity. I mean, we had a thing like, many years ago where our tagline was like, “beards of experience”, you know on the site, and we just had like, a picture of one of my co-founders friends, his name was Seth, and he had a big beard, on the site, and we got in so much trouble on Twitter. Like, “You guys are alienating female programmers,” and we were like, “Dude, we just thought that it was a funny tagline, and we put our friend’s face up with a beard.” 

Massi (00:12:49):

Yeah, same. Yeah, exactly. Same goes for us. It’s just a situation where you need to, yeah. I mean, it’s a sensitive topic, and you know, I understand both (Teja: Yeah.) sides. (Teja: Absolutely.) It’s just something like, yeah, you know, when you first start, you don’t really think too much about like, how that’s gonna look like in the future if you grow. Just like, we found it funny. We were like, “It looks cool.”

Teja (00:13:20):

And like, logo design is like, less than 1% of the things that you’re thinking about when you’re first starting a company, (Massi: Exactly, yeah.) you know? You’re just like, “Cool. That looks cool. Move on,” <laugh>. (Messi: Exactly, yeah.) So Metabob, so what does your company do?

Massi (00:13:34):

We automate the cloud review and debugging process using AI. So the company started when I was initially the EIR at a lab at Princeton, the city, from NEC, sponsored by NEC, the Japanese company, and I was working with our current co-founder and director of AI, got 40 plus years experience in NLP and different AI techniques. And per se, we were working on a model called, graph neural type of a technical graph neural net, or GNN, which was pretty new, like, initially published in 2017. And for us, we were looking at it for the specific applications on falco detections, and you know, the refactoring debugging code review space, there are several benefits to that. 

Massi (00:14:32):

We were actually comparing it with LLMs at that time (Teja: Mmm <affirmative>.) to see, like, we were also looking at for co-generation, itself, right? And kind of like, what we saw as one of the main benefits, is really the ability to understand context within the code. First, obviously, a graph. It’s able to read content like, content at its natural state, which is a huge benefit, instead of reading it as texts and represent it as a graph as well. And second, a graph, it’s able to see and understand different components within the code, even though they’re not connected with each other, and they’re semantic markers. So that’s another aspect that we found very valuable when we first started. So really, our bug mechanism, how it works is like, it’s based on analyzing the code with the GNNs, and the GNNs are trained on a large data sets of code with the known bugs and then learn to identify patterns within the code property graph that corresponds to potential issues.

Massi (00:15:43):

And then when presented with the new code, Metabob creates a graph representation of the code and applies the trained GNN to analyze it for similar patterns, right? And this is really possible to all programs, being fundamentally represented, the direct programs, and direct graphs, sorry, and that’s really like, the main value of GNNs. And so we decided to kind of, we first focused mainly on publications, so it was a very new space, and (Teja: Mmm <affirmative>.) we looked at, read a lot of papers and reached out to some of the people who wrote the papers in the beginning, as well. So one of our lead researchers was kind of one of the pioneers in the space, as well, and we started working together, and so we started again. The first couple of years was really like, publication focus to explore the space to see if that was a good application of it, because there are also other sides that LLMs are better.

Massi (00:16:50):

And so we were just comparing the two, and after a couple of years of, you know, just research, customer discovery, we actually thought, okay, there is a great opportunity for the main reason application I was referring to, and we decided to actually start a company. So we built an MVP. That’s where really like, I reach out to some of my open source friends to work together, build up a small team, and in couple of weeks, build the first MVP on GitHub and started getting some traction. And then that’s when we raised the first round, was in early 2021 like, mid 2021, and (Teja: That’s cool.) officially started the company. The unique value of the tool is the type of problems it’s able to detect. So, you know, to give you some context within like, the debugging or refactoring space right now, there are mainly two types of companies that are the more common linkers or static analysis tool, right, that developers or companies use as a first check.

Massi (00:17:50):

Those are rule-based, so they create a rule that, to identify problems based on similar patterns, right, (Teja: Right.) and those are mainly used for like, syntax or stylistic type of problems within the code, (Teja: Right.) so it’s kind of a first check. Myself, from a developer standpoint, it’s something that I’ve never really liked, and both from companies I work with and myself as they just generate so much noise, right? Meaning like, in 99.9% of cases, it’s kind of pointless the detections they give you, so you end up not really looking at it at all, because it’s more the time you go through all the problems to actually anything that finds valuable. And then obviously, nowadays, there are like, LLMs like, co-generation tools like Copilot Chat, which was, I mean, specifically Copilot Chat in VS Code, which was released in May. They do offer like, they’d say debugging, per se.

Massi (00:18:58):

However, the problem with LLMs with debugging is one, LLMs are not very good in constructing context within the code, and secondly, you always need to prompt it, right? So it’s something where, if you don’t know the problem, per se, if you just ask, “Oh, find me a problem,” it’s gonna find something, but it’s not gonna find you necessarily a bug. And if you ask the same things again, it’s gonna find you something else, right? So (Teja: Yes.) it’s an LLM, per se, or when it comes to like, definitely a better use case for co-generation than for refactoring or debugging your code. And actually, in a lot of cases, if you use Copilot, you’ve noticed that code generated with Copilot actually creates quite a lot of logical bags or security vulnerabilities. So what we’ve seen a lot of, our users are actually using us, along with Copilot to debug the code that Copilot creates.

Massi (00:19:59):

So going back to like, kind of our differentiation is within the, we’re kind of, so there are currently, those are the two other alternatives for us. We use our methods specifically to identify logical or contextual type of problems within the code. So we have a weird threshold, and our model enable us to find problems that are based on like, algorithm efficiencies, or you can think of all type of race conditions or ledge cases, memory leaks, or web framework problems, or let’s say, on Python, you use <unintelligible> any type of like, data type errors, data type errors or performative issues <unintelligible> so those are all the type of issues that we’re able to find due to our technique. And those are usually the ones that really like, take longer, right, for developers to figure out why those occur. So that’s what we’re now kind of, we focus on, just to provide like, to detect issues that are based on logic and context, and then we provide our AI generate recommendations to that.

Massi (00:21:16):

So we, our tool is pretty straightforward to use. Like, you can use it either on your ID or as part of the CI/CD, and after you add it like, every time you save the code, you run some analysis, we flag the problems, and we do both like, bugs or areas that can be improved, in terms of like, performance, and we flag it, and we give you like, without any input, and then we have a chat box you can ask. So we, our air creates like, explanations of like, what the issue is, and you can interact with it, so you can ask different questions to rephrase it, and then if you want to just apply, you ask for recommendations. We create snippets, you apply it, that’s it. (Teja: Mmm <affirmative>.) When it comes to recommendations, we do use another LLM for it, because again, LLMs are better when it comes to code generation, itself. But what we use with our GNNs is we create a context stream that then we send to LLMs. We build our own LLM in-house, but we also enable users to use other LLMs if they want to see if they use co-pilots or any, really any LLMs that they want to, and we just send the context stream to those.

Teja (00:22:30):

So what was like, the like, how did you guys first like, link the use of GNNs to code review? And it seemed like, at the time, everybody was using LLMs, right? Like, how did you guys make that connection?

Massi (00:22:48):

Well, I think it was like, as part of our research, it was really like, we looked at the what GNNs are, right, and what are the best application for it? Are they used to map different correlations like, in different industries, like, bioinformatics to like, map molecules, or…they’re used in different areas, right? And it was really like, obviously like, our expertise were mainly in-depth tools and like, our like, knowledge as well, when it comes to like, a space. Both our core teams, that was their area that we knew best. And so it was definitely a mix of like, I mean, on one end, it was the most rational like, part of it, so really looking at the value, the pros and cons of GNNs and what they can be used for, and also we did a lot of customer discovery, right?

Massi (00:23:46):

So I talked to several friends or people I know. Like, we’ve done, I don’t know, hundreds of customer interviews in the very beginning where what we did is like, I just reach out to engineering managers or like, CIOs like, product or project managers just to figuring out like, the way each company does code review and what was like, some of the biggest concerns, right, on their end, when it comes to like, obviously dev tool are pretty high when it comes to like, the cycle. And so figuring out like, what was some of the biggest problems there. And obviously, whenever you start a company, you wanna focus on a mass step problem, right, or (Teja: Right.) like, a must have solution to a problem. (Teja: Right.) And so on a certain level, like, obviously like, we found a different type of feedback in different industries, but we thought that was a strong case and based on like, the alternatives in the market, we also thought it was good timing, because a lot of new research was coming out, and at the time when we actually started, there wasn’t, I mean, in the past few years, you know, the like, ecosystem of that tool has changed a lot <laugh>. Like, (Teja: Yeah, that’s true.) it’s been pretty crazy in the past four or five years.

Massi (00:25:13):

So again, we first started, it was a different environment, and we knew that it was like, about to change, and we thought it was a good time. And so it was a mix to answer your questions, between customer feedback and just looking at the value, the process concept, the technique itself, and what was the initial use case. We thought that could have been like, it was a good area to focus, due to competitor, customer, and overall technique.

Teja (00:25:47):

For like, the lay person, of which I’m one, you know, maybe it’d be helpful to describe the difference between how an LLM works, and how you would train an LLM, versus a GNN, and if there’s any difference in the way that you would kind of construct these and utilize these for commercial purposes.

Massi (00:26:06):

An LLM, per se, are obviously deep learning models that use this like, billion parameter and an attentional mechanism to predict the most likely token (Abbey: Mmm <affirmative>.) to mainly to follow given input, right? While GNNs, we use an attention mechanism that comprised of both semantic and relationship markers that results usually in a more complete representation of the inputs. So usually like, the process we follow is like, first the GNN, the text and classify like, a problematic card with contextual understanding. So we train it. LLMs usually train on…the difference is like, our model, we trained it on verified code changes. So when we first started, we looked at like, GitHub or different open source projects, so high quality open source projects. We didn’t train it on the code, itself. So what we do is we look at code changes and verified code improvements, and we look at it like, why does that occur? (Teja: Mmm <affirmative>.)

Massi (00:27:10):

So that’s how we train the model in the first place, and then also over time like, we have added, obviously look at Stack Overflow, Reddit, and we’ve done a lot of company SOPs, so company design standards and annotations. Also depends on the language, right? For languages like Python, open source enterprise version is almost the same. While for like, Java, it’s completely different, right? So if you’re just trying to turn open source, it wouldn’t be good. It wouldn’t like, really apply to enterprise, and it’ll just have different applications. So that’s how we have trained it in the beginning. So, and all of that again, it’s then fed  to our GNN to learn kind of the best programming practices and to be able to map specific areas of the code, based on the surrounding context, and then we like, feed that.

Massi (00:28:09):

We create a context stream that we use for creating recommendations. So we sent to LLMs per set for providing the recommendation to that. So again, when it comes to pro and cons, LLMs in general are very good when it comes to like, because of the way they function, based on user input, they can, they’re good to predict a more likely following, which means like, their best application is usually for code text generation. (Teja: Right.) Well, when it comes to GNN, so again, Metabob, we create a graph representation of the code. The positive, the best side is one, we’re able to read the code in its natural state, (Teja: Right.) because again, all programs pretty much are fundamentally representable as a direct graph, right?

Massi (00:29:10):

Within each node of the graph, we specifically, we encode information about the section of the code base and the edge encode when it’s used, how data flows through it, and where it’s located within the context of the nodes in the code base, per se. So what that means is like, our technique, the main advantage is really like, understanding, being able to read different components within the code, understanding the relationship between different components within the code that are not connected to each other, right, and really like, and that’s the main value of GNNs in general, trying to map. And even though, (Teja: Got it.) again, those components are not connected or the semantic markers are not connected, trying to figure out the relationship between those through a graph, well, LLMs don’t do that. So again, GNNs are very effective when it comes to context understandings or, yeah, understanding the logic of it.

Massi (00:30:18):

Well, right now, the use case is not as proven on when it comes to generating new code. So that’s why we actually use LLMs for that side, right? And so the difference, to cut it short, because I feel like I’m <laugh>  giving like, a way longer explanation than needed, but it’s really like, when it comes to use case, LLMs call generation or text generations the “DI use case”. Well, for us, the DI use case is like, right now is back in the factory. So understanding what’s the logic of the code, and why specific problems occur, or how a specific area is not performing as needed, right? (Mmm <affirmative>.) So those are the differences right now. So we don’t really compete right now with Copilot or (Teja: Right.) any LLMs, per se. It’s not our intent to compete as of right now.

Massi (00:31:17):

Obviously, in the future, there is a potential when it comes to core generations as well, mainly using GNNs, but as of right now, we compliment LLMs. That’s really what we’re planning and what we’re doing. So a lot of developers or companies are using us along with LLMs one, to provide better results when it comes by using LMM, so better code. Because again, due to our technique, we create the context stream that can be used for both providing better code generations as we have better context understanding of the code that we can. So we create inputs for co-pilots, for instance, if you wanna do code generation or they use our technique for the bug or to identify the tech bugs and then LLMs, per se, to create recommendations to resolve the issues.

Teja (00:32:15):

So because your company is like, basically, you know, working at the frontier probably of GNNs, has it been difficult to scale the engineering team, or are you guys able to find like, good talent pretty easily? Like, what’s that like as a CEO?

Massi (00:32:32):

In the very beginning, what we have done is just reading papers and reaching out to people that wrote those papers.

Teja (00:32:40):

Oh, that’s cool.

Massi (00:32:41):

I spent a lot of time just to like, trade notes with different researchers at universities all across the world, really, like, from Toronto to Europe and Amsterdam. So it was like, that usually has been my approach when it comes to the first stage. Whenever you come up with something, you are in the forefront of research, right? You want to train us with the other few people that have been in the forefront with you. So (Teja: Yes.) They have great feedback, and what I’ve noticed is, in many cases, if what you’re doing is valued, you may actually find, you know, coworker hiring through that process, right? So that’s really what happened to us when we started. It was me, Ben, and Avi, and we spend the first few months just reading papers, reaching out to the researchers that wrote it, getting into calls, trade notes, and that’s how we found the first couple of like, partners or like, researchers that work with us.

Massi (00:33:51):

And we got a few grants as well, related to that. So working with researchers, obviously, they have great experience in writing grants, and so that’s really what, before even raising money, what was our approach. (Teja: Yes.) So when it comes to like, the R&D side, that’s what worked for us. When it comes to the more like, day-to-day developments, (Teja: Yeah.) I really tapped into my network of open source contributors. That’s what I’ve done. Usually, I found that to be the most effective, because, in terms of quality and price, right, (Teja: Yeah.) and, yeah. So that’s helped a lot in the past, like, landing this company to just find people that I know already. They have a lot of things I can look at when it comes to commitment and other expertise, and I found good rates overall. So they enable us to keep a pretty low burn rate or good quality, because hiring, that’s like, going back to like, my previous companies, it’s been, it was very tough in the beginning. (Teja: Yes.) So that’s definitely something I learned over time, and it’s one of the most important parts, right? To make a company successful as you grow, like, finding the right people. (Teja: Yup.) That’s that’s kind of been our approach to it.

Teja (00:35:19):

Yeah, it’s so interesting, because it’s like, when you hire, you can potentially bound like, the scope of solutions that you have in front of you if you make the wrong hire, and then like, it’s like, and that can compound. They didn’t hire people, and like, shit, you know? What are we doing? Cool, that’s awesome. Ten years ago, did you think that you’d be running a tech company? Like what did you think you’d be doing today?

Massi (00:35:45):

Well, ten years ago, I probably would’ve thought that, yeah. (Teja: Really?) Maybe 15, 20 years ago. Yeah, I think so, because like, it’s when I first moved here, and I would say like, when I probably my like, 18th, 19th is when I first like, moved from like, it became like, from just a hobby to something I kind of wanted to pursue as my, you know, there’s nothing better in life, right, than doing what you love. (Teja: Yeah.) So I was like, let’s just try to make it a business, right, so find something that actually people use also. It’s very rewarding, I think when you build something that people like and give you good feedback, I think that’s a really what makes being an entrepreneur like, it’s the best part of it. Like, seeing people using what you have built. (Teja: Yes.) I would say 10 years ago, I was already, like, I kind of, I mean, I hoped that was gonna be the case, for sure, (Teja: <Laugh>.) but yeah. When I was younger, I don’t know, I wanted to be a, like, an astrophysicist, so it was completely different goals in life, but I guess that’s a change.

Teja (00:36:58):

Do you find like, your days these days being…like, are you still able to program or do you find yourself doing more business stuff these days?

Massi (00:37:07):

It’s probably 20/80 right now. It’s like, it’s been shifting more and more towards business activities. We have a fantastic team of developers that are better than me in a lot of areas, right? And so you always wanna be up to date anyway with like, new techniques, new frameworks. I mean, I still do a bit of the development, for sure.

Teja (00:37:31):

Are you embracing the transition to becoming like, a full-fledged, like, CEO businessman? Or are you trying to hold onto as much programming as you can?

Massi (00:37:45):

I don’t know. It depends on the day. (Teja: <Laugh>.) It’s hard. I can’t give you like, a yes or no answer. I guess it’s, some days I will want to just to like, you know, just program and like, do that. Other days, it’s exciting to be a CEO. Travel, we do a lot of conferences, a lot of competitions, and I mean, I love that part of it, but you know, when you are a CEO, the two main things you do every day, it’s like, fundraise and sell, right? So (Teja: Yes.) I mean, I don’t dislike it. I think it’s cool. Especially like, I like actually the business development side more than the fundraising side, for sure. It’s the fundraising, per se, it’s, yeah. That part I will like, happily avoid if I could (Teja: <Laugh>.) but that’s the, you could ask for money. I guess that’s just, that’s what you do as a CEO, really.

Teja (00:38:46):

Yes. I’m right there with you. I mean, we like, bootstrapped the company for a couple of years and just kind of grew it outta revenue, and then we raised our first round, I mean, first institutional round, we had some angel money, you know. The business changes when you have professional investors, (Massi: Yeah.) you know? And I think I remember how it was before, and I know how it is now, and it’s very different, you know? (Massi: Yeah.) I hear you on the selling. That’s my favorite part, too. I love selling, and recruiting’s fun, too. I’m sure you do a lot of that, too, like (Massi: Yeah.) trying to get on new people, yeah.

Massi (00:39:22):

Yeah. Yeah, that’s fun, for sure. Yeah, business development I enjoy a lot. But, yeah. Well, again, as a CEO, I guess it’s a fancy word, but really what it means, you just beg for money all day. So that’s what you gotta, you learn to do it and just embrace it, I guess, just until, which is, yeah. This company, you know, it’s in a space where it’s hard to be profitable very early on. (Teja: Yeah.) The same as you, like, my very first venture, like, we raised a tiny bit first, like, again, angel money, and Europe, the ecosystem of investors is different, so (Teja: Yeah.) it’s harder to get like, a lot of money (Teja: Yeah.) at a very high valuation unless you, I mean, pretty much, in any case, but especially a seed or pre-seed.

Massi (00:40:18):

So we actually were able to get to our revenue side, like, almost profitable very early on. So it was like, more raising just to push growth. But when you’re working with AI, it’s hard to, especially in the beginning, you have so much cost from, you know, GPUs to cloud cost. Like, that’s obviously the biggest challenge of any like, company within the generative AI space, really is to maintain and like, allocate resources in the early stage, right, without going broke. And for us, we have a free tool, and we, our tool is free pretty much for all the ideas, and so most of our users are using us through the free tier, and then we have an on-prem solution for enterprises that can be run locally. That obviously doesn’t cost us anything, but most of our costs for supporting the free tier, which is obviously something we wanna keep pursuing as, again, just to support the open source and getting feedback from developers, I think is key when you have dev tool. (Teja: Yeah.) Yeah, because of that, that definitely represents our biggest cost right now. Just cloud cost in general.

Teja (00:41:37):

Yeah. Cloud cost, GPU, even the labor cost of getting really good people, and all of that’s front loaded before you have a thing to sell. That’s, yeah. That’s just the nature of building something at the frontier. But I hear you. It’s funny, like, when we first started the business, I actually just spent like, eight hours a day going down angel lists and just cold calling companies, because I was like, I don’t wanna ask anybody for money, we’re just gonna see if we can get some sales. (Massi: Yeah.) And then after a while, you know like, okay, we need to have some people do this. This is crazy. So <laugh>, it’s like, (Massi: Yeah.) yeah, yeah.

Massi (00:42:16):

That’s what you gotta do, I guess. Yeah.

Teja (00:42:18):

That’s what you gotta do, man. Yeah, totally. As you look like, two to three years ahead, you know, what’s up next for Metabob? Like, what you guys working on? What you guys wanna bring to market that you can disclose, you know?

Massi (00:42:32):

Yeah, so well right now, so we spent the first couple years since fundraising to really focus on our free tools. So we, initially, we were on GitHub, then we just heard developers. They wanted to have the tool on the IDE side, and it’s just more interactive, right? We saw the trend, everything is moving towards left in the development cycle. So we now are available on ViaScope, and we’re moving more, to more and more ideas, and then [in the] next few months, we’ll be available for most of those. We’re also, we started with Python only, and right now, we’re gonna release in the next couple months TypeScript, JavaScript, and C, (Teja: Oh, nice.) and support the languages, and we’re already working on more languages support. Our model, per se, is language agnostic, so we have to do, following up, unsupervised model where all we have to do is like, look at the categories, the model identifies for each language, and pretty much, it’s not like, a label in itself, but like, we can just like, put a label in the category, per se, to make sure there is enough differentiation. So it’s usually a pretty straightforward work process toward new languages, and we do that for enterprises, but since April this year, we started to like, monetize the tool, so we launched an on-prem solution. We have already a couple of Fortune 500 and (Teja: That’s sweet.) over 50 company in our SaaS.

Massi (00:44:06):

So we have like, an on-prem side. A big problem that we’ve seen in the market right now is related to legacy code. So a lot of companies that have dealt with or have legacy code and complex cloud base, obviously those usually, they’re hard to update, and like, they always lack documentation, and as time goes on, it becomes a bigger and bigger cost. Actually, we heard from companies that sometimes maintenance of legacy code costs almost 70% of their IT budget. (Teja: Mmm <affirmative>.) So what a lot of bigger companies are using our tool for is to be able to, for maintaining that, the quality of it, and refactor legacy code, our tool and other great like, benefits of GNNs is, it can be easily customizable. So we train the model on the customer and notations, so the review code history, and that provides way better results, right, for enterprises for longer cut.

Massi (00:45:09):

So in terms of like, the next couple years, that’s definitely a market that we’re penetrating more and more. And so we’re definitely, I can see the shift a bit more to like, just offering a free tool to get developers’ feedback, which has been our like, initial objective, following a bottom up approach, but now we are to a point that we keep getting new referrals. So the bottom up like, is finally like, it’s definitely now getting us to a point where we’re growing organically. We still don’t really do much outreach at all. It’s just through organic leads, which is really what we wanted when we first started. (Teja: Yeah.) Yeah, in the first couple years, like, we’re gonna focus mainly like, adding language support. On a soft side, we’re getting, you the SOC 2 compliance certifications that really help us also to boost our SaaS.

Massi (00:46:12):

But definitely, that’s gonna be our focus to grow our enterprise customer pipeline. And in terms of features, we are just adding more categories, and it’s always been a goal for us to be also as part of the code generation side. That’s more like, a longer term goal, but combining GNNs to LLMs, it’s what we see is the ideal scenario for code generations. And we’re, I’m sure it’s going to happen. I’m sure the big players are looking at it right now, and we’ve seen the past year or so, huge like, hype around GNNs for applications. Companies like Amazon are now using it. And combining the two, I think is ideal for code, because again, the GNNs can provide great context understanding to create the context stream to be used to then generate code. And so we think, either us implementing, focusing more and more to develop our own LLM, along with the GNN technique, we have developed to do also co-generation or partnering up with one of the bigger players for LLMs to do so. It’s definitely in the cards for us for the next couple months, for the next couple years probably, and so (Teja: Awesome.) that’s the direction we see us going.

Teja (00:47:45):

That’s awesome. That’s exciting.

Massi (00:47:48):

Yeah. Another great support we got, it was like, we started as part of this accelerator, sponsored by NEC. As I said like, I was working as [an] EIR for NEC and partnership with like, a researcher from NEC labs at Princeton. And yeah, so that’s actually helped us to go through the customer discovery part, right? Like, they were very good, in terms of like, giving you a framework to test your assumptions. So because the technique we were using was pretty new, and we were focusing on coming up with use cases for it, I think the key, it’s always to test your assumption with potential customers. And so we went through like, almost a year of customer discovery, like prompted by the accelerator at NEC X, which really, their goal is to match researchers with entrepreneurs.

Massi (00:48:42):

And so I was working, again, my role as EAR was working with the researchers to test also the technique we were developing. And so that has been helpful for us, for sure. That was key to also be able to apply the technique to refactoring and debugging, because thanks to the customer discovery part, we directly heard from engineering managers and so on, with over 500 interviews to really get that, validate the need, right? And that really was what prompted us to start the company and to make it happen, right? Those two years from 2019 to, I mean, it was 2018 and 20, I guess, where we went through NEC X Accelerator and then Alchemist. So I definitely recommend always, when you start a company, accelerators are always good. I think they give you, not just mentorship, but if you find the right accelerator to get the right connections and like, not just the framework, which usually, the framework, you know, you can find on like, YC school, or there are a lot of resources, (Teja: Right.) but also just to give you the right intros and trade notes with you, get the different perspectives, and then obviously, demo day is always great. But that for us was very helpful. Like, especially when, because we started working with researchers having like, the NEC X side to give you that framework to like, and help you out to arrange customer interviews. And so, it was valuable. for us, for sure.

Teja (00:50:25):

Shout out to NEC X.

Massi (00:50:27):

<Laugh>. Yeah. You said you’re based in Nashville, right? Tennessee?

Teja (00:50:32):

Yeah yeah, Nashville, Tennessee. Yeah. (Massi: That’s cool.) Yeah, you guys gotta come visit. I mean, there are a ton of companies here now, so if you guys ever find yourself here for a conference, you need to come through, for sure. I’ll take you out for a beer.

Massi (00:50:48):

Yeah, I’ve heard great things about Nashville. I love music too, so there is a great scene there. I’m definitely super down. I’ve never been there before.

Teja (00:51:00):

Seriously. So do you like live music? Like, country music, that sort of thing? (Massi: I do.) Okay. Hell, yeah.

Massi (00:51:07):

Yeah. Actually, I was just, we were panelists at South by Southwest, so it was my first time in Austin, and Austin is also great for that. (Teja: Yes.) It’s like, they have a fantastic music scene. Yeah, I’m used to the Bay Area, which it’s quite boring, so (Teja: <Laugh>.) it’s like, I’m always down to travel, right? And so there is great, you know, great culture here, great, super, like, bright people, but like, when it comes to like, I don’t know, nightlife or just like, social life, it’s not where you wanna be.

Teja (00:51:43):

Hundred percent. And I, every time I visit the Bay Area, I’m like, “Man, these people are like, too smart,” and I get kind of like, I wanna be around people that, this sounds bad, but I wanna be around people that like, don’t do what I do, and like, I can talk to them about like, other stuff besides your job and like, what you’re trying to build, you know?

Massi (00:52:04):

That’s the thing right now, yeah. (Teja: Yeah.) I fully get that. It’s like, I mean, it becomes a bit overwhelming, right? So yeah, it’s something I’ve noticed myself, too. So it’s good sometimes. For me, I’ll be going to Europe in couple of weeks, and that’s a different life, right? (Teja: <Laugh>.) Like, people in Italy or like, part of my family lives in France, so it’s just like, at least like, you go there, and you like, it’s a different mindset. People live to like, enjoy life. They work to enjoy life, not to like, they don’t like, just live to work, right? While people here, it’s, I mean, there are pros and cons to both, but it’s good to have a nice balance.

Teja (00:52:47):

And France is like, you know, you work three hours, take a two hour break, work maybe two hours, and then you’re chilling, you eat your dinner at 9:00 PM, wake up next day, go to work at 10, it’s four or five days a week. You know, here, especially in the Bay Area, you know, it’s like, if you’re not working 12 hours a day, people are like, “Are you okay? Like, what’s wrong?” You know? Right? (Massi: Yeah.) It’s very different <laugh>.

Massi (00:53:12):

Yeah, of course. Yeah. I mean, in Italy, it’s, yeah. Again, there are pros and cons. I think when I (Teja: Of course.) moved here, I really liked the, like, the opportunity, and you have so much more like, ambition and drive to do things, to actually get things done. But then again, after 10, 15 years in the Bay Area, (Teja: <Laugh>.) we’re like, we go to Italy or like, Europe, and you know, you just enjoy life for a bit, and you’re like, “Damn. That’s living.” Right? But then again, if I stay too long there, I become like, I kind of wanna get into like, I wanna be more active in terms of like, work ethic. Think when you’re in like, 20s, 30s, that’s really the time to hustle and get things done. So it’s good to have a nice combination of the two. That’s always the answer, right? Nice balance.

Teja (00:54:02):

Totally. I spoke to this CEO, she was, maybe she was based in, I think the Netherlands, like, in Holland, I asked her, I was like, “What’s it like building a company in Europe?” and I could tell that she was like, a capitalist and like, very driven. And she was like, “It’s annoying when people take two months off in the summer.” I’m like, I can only imagine what that’s like (Massi: Yeah.) if you’re trying to build a company, you know? It’s like…<laugh>.

Massi (00:54:30):

Yeah. Especially in like, August, it’s like, pretty much like, Italy’s closed in August. So (Teja: <Laugh>.) you go there, like, you won’t find any Italian, really, like, aside from like, the beach. (Teja: <Laugh>.) You go to like, big cities, it’s like, “closed for business” signs everywhere, pretty much. So that’s, (Teja: <Laugh>.) yeah. But again, if you are living there, it’s like, why not? Right? Like, you live once, and if you can, just like, enjoy life. If you, you know, people there are a different type of people, and some are just like, they’re like, “Okay, I work to pay my bills and stuff, but then the rest I wanna just enjoy it,” and hey, kudos to them, to be honest. If I could do that, I would be very happy in Italy.

Teja (00:55:27):

Totally. The problem, and I’m sure you’re similar, is that like, for me to have my mind stimulated, where I’m not bored, all the jobs are very demanding. Like, everything that’s intellectually gratifying just requires a lot of you every day.

Massi (00:55:44):

Exactly, yeah.

Teja (00:55:45):

And there’s no way, you know, back, I mean, even in India, it’s the same thing. I found that there was no way to get sufficient intellectual stimulation, you know, if you work a government job or something like this. (Massi: Yeah.) It’s a very relaxed lifestyle, but you’re kind of like, going crazy a little bit.

Massi (00:56:03):

Yeah, yeah. I will never be able to do so. Like, when I actually, when I was at my first company to like, pay the bills and stuff, like, I had, when I was like, 16, so I had like, I tried billions of different jobs, because I was already traveling and trying to like, bootstrap it, and I worked as a bartender, waiter. I worked in a warehouse. I literally worked in a warehouse, like, moving boxes and like, boxes. It was this warehouse where [we were] like, packaging books, like, school books, (Teja: Yeah.) for like, and send it to schools, like, huge amounts. So it was midsummer, and I was just by myself like, carrying billions of tons of books, (Teja: <Laugh>.) to pallets and shipping them. It was hot. I picked up fruit. That was like, one of my jobs, literally during the summer. Like, go to like, [the] field and pick up strawberries and stuff. So I tried everything, you know, like, (Teja: Yeah.) that was more like, have to find things to do, because obviously my parents were like, “If you wanna like,” (Teja: A hundred percent.) “to be an entrepreneur, you gotta, like, we definitely can’t sponsor that.” So I just like, (Teja: <Laugh>.) try to like, make some money on the side, because when you started as an entrepreneur, especially in Italy, like, you make zero money for quite a lot of time, right? So I had to, I tried everything, but at least I always had like, the mindset. I’m like, “Okay, I’m doing this as like, to be able to like, make my dream come true.” But I couldn’t see myself like, working for, as you say, like, any type of like, institution that you’re just like, every day, you wake up, you do the same thing, and then you go back home. Again, some people can and great. Kudos, because it’s like, it’s great, I think, great like, mentality balance. When I was younger I was like, more judgey towards that, but now, (Teja: Yeah.) I’m like, “Hey, I actually, I’m jealous,” right? I respect it a lot, because it’s like, if you can just enjoy life, and see work, and just, you know, spend a couple hours a day just to do that, but then your life is actually spent with your family and like, enjoying the small things. It’s actually something I wish I could have as well. Like, I just, my mind is not wired for that. It’s just, I become so bored so quickly, every time.

Teja (00:58:29):

Same. No, yeah. I often think that, but you know, it’s like, in a like, a society, like, if we were in like, a primitive society, like, you need like, there’s a reason why our minds exist today after millions of years of evolution. It’s like, there’s a role that we play in society, and we probably used to play back then. Like, we’re the people staying up at night thinking about random shit. That’s, you know, and like, trying to perfect some small spear to kill, you know, an animal better, you know? And we’re not the people that stop hunting at 4:00 PM, or whatever. It’s just how it is, you know <laugh>?

Massi (00:59:07):

Yeah. To each their own, right? I think it’s like, it’s a good combo. You can’t have only people that’s like, otherwise, like, we will not function. And so (Teja: Yes, yes.) I think it’s good to have a mix. One last question, I know we’re running outta time, but why, you ask the name of why Metabob, why is Gun.io? Where does that come from?

Teja (00:59:31):

Back when we like, first started, we were like, a bounty site for open source projects. And so we were like, “Hey, you could like, hire a hired gun on the web to like, hit this open source bounty.” So that’s kind of how we started. And we, you know, if you look at our site, and if you look at kind of the branding in the app, it’s all about like, space cowboy. Like, I don’t know if you watch anime, but we like, Cowboy Bebop.

Massi (00:59:59):

Of course I do. I was gonna say.

Teja (01:00:00):

Okay. And so like, we like the whole theme of Cowboy Bebop, we like western shit, and so that was kind of the original theme of the company.

Massi (01:00:09):

I like it. It’s very cool.

Teja (01:00:10):

Thanks, man. Yeah, and trust me, we’ve gotten the whole thing around like, “Gun? That’s not good,” and I’m like, dude, it has something to do with like, a gun. Like, we’re not selling guns, we’re just, it’s like a cowboy/cowgirl theme, you know?

Massi (01:00:24):

Yeah, yeah, of course. Yeah. You gotta also like, learn how to like, filter different comments, right? Like, if you listen to everything, then it’s like, you can’t move on. It’s like, again, there is sometimes too much of that in my opinion. So you guys are selling like, man, you’re really like, just connecting talents to companies, right? So it’s like, (Teja: Yeah, yeah.) I don’t see anything controversial there <laugh>.

Teja (01:00:51):

No, totally. Yeah, and the devs love, like, you know, the devs that use our platform, like, they like being like, kind of professional, you know, mercenaries, you know, to go and do a project. It’s cool. It’s a cool kind of like, motif. I have an office in this house, but I like, need to like, make it kind of like, a western theme. So I just, I like the Wild West, and I like the Frontier, so that’s kind of why we did it.

Massi (01:01:21):

Well, I like it. So (Teja: <Laugh>. Thanks, man.) It doesn’t matter too much, but hey, you get one supporter, yeah.

Teja (01:01:27):

<Laugh>. Yeah, because you know, I kind of feel like, half the fun of like, building a company is you get to define like, your own company’s lexicon, and you have your team, you know? That’s my favorite part. So where can people find you and Metabob the interwebs?

Massi (01:01:45):

You can find us, obviously, Metabo.com. We are on VS Code Marketplace. So again, the tool is free for any developers who would like to check us out there, give us feedback. We always love to hear that. Right now we are, the tool is Python only on VS Code for the free tiers, but in the next month or so we’ll add the JavaScript and TypeScript support, and C is coming as well in the next couple months. And you can also find us on GitHub, same things. GitHub Marketplace or Bitbucket and GitLab. (THE FRONTIER THEME FADES IN) So obviously, most of our users right now for the free tier are using us on VS Code. But if you have a small company, you’re interested, check us out at GitHub Marketplace or schedule a demo through our website deployments, and we can like, give you a demo and tell you more about the company.

Teja (01:02:40):

Awesome. Awesome. Well thank you so much for your time, Massi.

Massi (01:02:44):

Thank you so much, Teja. I really appreciate it, and I hope you have a great rest of the day.

Faith, via previous recording (01:03:00):

Thanks for listening to The Frontier podcast, powered by Gun.io. We drop two episodes per week, so if you like this episode, be sure to subscribe on your platform of choice, and come hang out with us again next week, and bring all your internet friends. If you have questions or recommendations, just shoot us a Twitter DM @theFrontierPod, and we’ll see you next week.

(THE FRONTIER THEME ENDS)

Watch: