The High Flyers Podcast with Vidit Agarwal
00:10:18 · John Willett, Vidit Agarwal
[00:10:18] S1: So confident. Like, so much just sense of presence, sense of self. Super funny. I mean, out the gate, just, like, hilarious. But not, like, warm and fuzzy. He was a little bit intimidating when I first met him, but always just, yeah, super smart and sharp and and hilarious. [00:10:34] S0: So how did you break in? How did you go from that to becoming friends and then starting companies? Right?
00:10:40 · John Willett
[00:10:40] S1: I think it was classes that we took together. I'm blanking on what the first couple ones were, and then, yeah, we ended up in in some of the same kinda, like, social clubs within Princeton. And then, I mean, the company thing, we did a little bit of independent research together when we were seniors in college. Well, really the story there was... So Gabe had taken some natural language processing classes and we started kind of chatting about it or, you know, junior years. And I had just finished my, like, junior paper for economics. And one thing we just kept talking about is there were all these people in the economics department at Princeton who, like, smart, but they're not technical or they're not... They don't know how to program. And the weird thing about these independent research projects was that you might spend like two weeks coming up with like a topic like, okay, I'm gonna do like an instrumental variable study on this and this. And then like, there's a bit of like collecting the data. And then a lot of it is just literally like writing the bar code or the Python code to like cleanly do the analysis, clean the data and do the outputs. And so it felt like 80% of the work that went into these financial or economic studies were things that the people weren't good at. Like, hated it. They'd like figuring it out from scratch. And anytime you see something like that, it just sticks out. Right? Like, this is a weird dislocation of resources. So we started looking at... Mean, I LLMs were around, but they weren't like that good and there wasn't like an API you could like ping to get like a good, you know, LLM response. And so a lot of it was kind of these more anachronistic grammar based approaches. And we were pretty impressed with a rule set that's not so crazy. You could take pretty open ended strings of text and convert that into more like deterministic like code runs. That was basically the the insight, which wasn't like a novel insight by any stretch of the imagination, but we were really interested in it. So kind of the idea that we set for the next year was like, okay. What if we can, like, build kind of a product that allows you to, as an economic student, just kind of describe what you want, say, hey. Here's this big clunky dataset. Can you remove the outliers and get rid of parts of the data that clearly are mislabeled?
00:12:41 · John Willett, Vidit Agarwal
Like, those sorts of things. Like, do the cleaning, do some of the exploratory analysis, and and then do the outputs. Just describe that and then the system could sort of piece together like a Python script to do that analysis. Again, like, this is all stuff that within a day of GBD three coming out would have looked so horrible, like did not work that well. But for the time, felt like it was gonna be helpful. So we spent a lot of time building that after our senior year, we had a few months before we started our, like, investment banking jobs, and we did talk a little bit about, should we try, like, commercializing this? [00:13:13] S0: Was it called Vogo? [00:13:14] S1: This was No. This was before... So we... The original product we called Athena, which is just, the most cliche possible, like, technology name. Like, every single product is called Athena. So from the jump, there was no way from a legal or branding perspective that that was that that was gonna work. But at the time, yeah. [00:13:30] S0: Was that... How different was that to the first Vrgo product that you eventually went with?
00:13:30 · Vidit Agarwal, John Willett
[00:13:30] S0: Was that... How different was that to the first Vrgo product that you eventually went with? [00:13:35] S1: Quite different. Quite different. Because the first one didn't incorporate LLMs at all. It was all like uploaded data and it was all like... It was more structured data. It was kind of like, oh, like I have these CSVs or like there's this like dataset I'm integrating with and you would do basically just like structured analysis on top of that. Whereas the first kind of versions of Rogo, we're actually much more like we're plugged into all the like market data and like company fundamentals and company financials and you can ask a question and it would basically just pull those public pieces of information. This was still like Faxat and S and P Global and LSEG still were not picking up the phone. So it all these like random like Databento, like these random third party market data providers that no one in investment banking would ever use or trust, but it was sort of more of a proof of concept. And then the first product was also like a command line interface. So it left a lot to be desired in terms of UI UX to say the least. [00:14:28] S0: It's interesting because when people think of Rogo today in August 2026, people think of Gen AI. Right? Yeah. Back then in 2019, like you said, was pre ChatGPT going mainstream, and it was pre AI disrupting software in general. Yeah. So I'm curious now with the benefit of hindsight when you reflect on that moment in 2019. I guess, what did you know and, like, what did you maybe misunderstand?
00:14:50 · John Willett
[00:14:50] S1: The latter category is much more than than the former. I mean, we, like, we had, like, vaguely an idea that this stuff was cool. Like, that... I think that's probably the extent of our knowledge. There was maybe also a vague sense of, like, this stuff is both cool and getting better. And there was some concept of like Moore's Law and it was all very vague. We were like, okay, maybe this stuff will get better. Had no idea of the timelines or where things were gonna end up. And to give you a sense, it's like we started our banking jobs. We did kind of somewhat quickly where we're like, you know what? We do think this stuff could be applicable to our like front office banking jobs. I'm spending my entire day pulling data from FactSet and then from Cap IQ, PitchBook, Crunchbase, Dow Jones, all these different data sources. And so it felt like some of what we had built could apply, but we had basically no urgency whatsoever to start the company. Like we were like, oh, maybe we'll do, you know, in five to ten years, we'll pool some capital and bootstrap this or start working on it at some point, which I think is a pretty good indication. Like, if we were smarter and really knew where the research was and and knew where things were going, we obviously would have just quit our jobs and started the company immediately. The reason we ended up doing that was our first investor was this guy Kevin Ryan who founded, yeah, MongoDB and Business Insider. And, he's kind of one of the best early stage investors certainly in New York. And he recognized it, I think, little bit before us, and and we were talking about what we were doing. We were like, hey, maybe we'll apply to like Y Combinator at some point. He was like, no, don't. I'll just give you $2,000,000. You should quit your jobs now and start the company. This thing's gonna move a lot faster than you think. So that made it a lot a lot easier and he he Wait. Helped
00:16:25 · Vidit Agarwal, John Willett
[00:16:25] S0: Can you connect those dots? Because by then you had already graduated from Princeton. Right? [00:16:29] S1: Yeah. Yeah. So this was like a year and a half into investment banking or something like that. [00:16:33] S0: And so you were doing this on the like weekends and nights? [00:16:35] S1: Yeah. Basically we're doing a little bit of... I mean, so like at this point, not that much of it was like actual company building. Like we, we were working like eighty, ninety hour weeks in banking. So there was very little like product or like selling. Like we were maybe like doing small user interviews and stuff, but very little. It was much more by the end, like kind of figuring out, like, defining the vision for what we wanted to build and actually putting together, like, a pitch deck, which seemed like a a very important exercise at the time to have a really long and comprehensive... Look, it was such a good investment banking style pitch deck of just, like, so detailed about all of our all of our thoughts. [00:17:09] S0: I I was gonna say exactly that when I spoke to your investors, they said that's the difference in like your board decks compared to other portfolio companies is that both of you worked in iBanking. So your Yeah. Like decks are like very like iBanking style. [00:17:21] S1: Oh, that's so... Because I would have thought that like now we've gotten away from that. So if they still think that, that you guys should see this first one. It was horrific. It was also... Like [00:17:30] S0: some founders write memos. Right? Was I think you guys still do decks?