TUM.ai, MIT and Y Combinator. Meet Alex.

Alex started seriously coding right before college and got into AI even later. Still, he managed to achieve impressive things such as leading TUM.ai, doing AI research at MIT and founding a YC startup

Studies BSc CS @ TUM ‘23, Bachelor Thesis @ MIT ‘24
Start-up stormy.ai (formerly One Interface) (YC S24)
Experiences AI Engineer @ Rohde & Schwarz, Data Scientist @ Lilio, AI Engineer @ Heimkapital (all TUM.ai projects)
Organisations President & various roles @ TUM.ai ‘23
Origin Ukraine
Links LinkedIn

---

Growing Up in Kyiv

Where did you grow up and what projects were you involved in?

I grew up in Kyiv, Ukraine in a very educated family that put a lot of financial and time effort into educating me and my brother. I started learning English at age three. Then I did piano, winning my first international competition at age six. After eight years and five more competitions, I made a big pivot to STEM because I wanted to make more money in the future. I went to one of the best public schools in Ukraine, did Olympiads in chemistry and physics, that sort of stuff.

Why did you choose to study Computer Science?

At school I was doing mainly Chemistry and Physics, they were interesting to me. Then I chose computer science, because I thought it was the path to have the most impact, unlike say theoretical physics where you research some particles for 10 years only for it to get disproven a year after that.

Studying at TU Munich

Why did you choose TUM and how was your experience there?

I actually applied to American colleges, a few in Germany and applied to a couple of colleges in Taiwan. The US was my first choice, I got into NYU and was planning to go there. But then my family’s financial situation deteriorated, so the only thing I could afford was TUM at $200 per semester. It turned out fine though, because if I went to NYC, Covid would've started one semester in and it would've been an awful experience.

What was your focus and approach at TUM - studies or side projects?

At TUM, I made sure to minimize the time I spent doing university related stuff. TUM is a big public university — the CS department alone admits like 1500 people every year. So naturally I wanted to do something to set me apart from those 1500 people per year. I probably spent around one-tenth to one-fifteenth of the time I was expected to on studies. It allowed me to do some side projects, like leading TUM.ai, getting into MIT for AI research and building my current startup. My goal with these projects was basically to become the best technical CEO I could.

Because I was doing other STEM stuff in high school, I started coding late. I actually started coding like 9 months before TUM to prepare myself. In the first two semesters, I was doing just TUM coursework. I was taking like double the courseload though, because it was Covid and not much interesting stuff can be done on the side. So I just used that time to take mandatory classes. And then when Covid restrictions subsided, I joined TUM.ai.

What do you think about TUM and university in general?

Before I get started, let it be on the record that I’m really grateful to TUM for giving me the opportunity to live in Germany and for only charging me 1500 bucks for the entire degree. But a lot of things at TUM are done very slowly and impractically, like taking weeks to prepare simple documents, jumping through hoops to get a nonstandard minor at a different university etc.

There’s also seemingly little to no accountability that professors have for how well they teach their students. 70% failure rates on some exams are a commonly known symptom of that. In general, it all feels like the students are there to detract from professors’ research, which is more interesting to them.

This is actually something people at MIT complain about too. However, some MIT profs go into teaching because they just want to do teaching. At TUM, I think I had exactly one singular professor that was like that.

The good thing about TUM is that it’s a huge public school. It has 1500 students per lecture for an average size CS class, and there’s no attendance taken. You don’t have to interact with TUM much to get your grades and eventually your degree. So I would do all my side things throughout the 14 weeks of the 16-week semester. Then on the 15th week, I would sit down, cram the whole semester's worth of information in a week, pass the exam. My grades are shit, but in CS no one really cares about grades, especially not investors in my company.

Growing and leading TUM.ai

How long were you at TUM.ai?

Two years as an active member, growing from a normal member to president, and then half a year on the board.

Why did you spend so much time in the TUM.ai student organization?

So the reason behind getting into it originally was to get the soft skills necessary to be a good tech CEO and set myself apart from other CS people. But while there, I worked on consultancy-style projects where we hired a team of students and developed a complete solution for a company with a huge degree of responsibility and freedom. It was really valuable because I learned B2B sales, hiring people, and marketing to prospective hires. Oftentimes in those projects, I was the youngest, like the only bachelor’s student and everyone else would be in their M.Sc. or PhD. And so it was just a great environment for me to interact with people who were way older, more experienced than me in AI. That really helped me to progress fast.

I consciously sought out people who were further along than myself to learn from them. That was my strategy once I got into TUM.ai. Also, it was amazing to see TUM.ai grow from 30 members and a couple events every semester to over 200 people, 200K income and 10 departments (each with multiple projects). We really rode the AI wave.

You also did a lot of internships over the summers. How was the experience working full-time?

Those were actually not just any internships, but every job you see on my LinkedIn is actually a TUM.ai project that I helped organize and sell to a company.

What’s it like working full-time?

I enjoyed the learnings that came with it, especially since I was always the youngest on the team and had people to look up to. I didn’t enjoy the amount of bureaucracy at some of the companies or the lack of communication some startups I worked at. I definitely learned some things that were well done, some things that were poorly done. And I’m going to apply this to my company.

Motivations and impact

What drives you and makes you different from others?

It’s probably the breadth of my interests, which is not a good thing per se, but it makes me think differently. I could go on for hours about obscure topics in history, linguistics, etc. This breadth takes time to develop and I’ve switched my main interest a few times, but it helps me stand out by thinking in non-standard ways. I make non-obvious good decisions that stem from the huge corpus of data I know from vastly different parts of knowledge.

In general, when it comes to learning about and researching different topics, I’ve kind of always been like that. I could say it stems from my childhood where my family was really encouraging me to learn a lot — no matter what exactly. And we would have all those long, interesting discussions about all sorts of things, drawing non-obvious conclusions from different parts of knowledge and making those inside jokes that no one else would understand. My little brother is also like that: shows my sense of humor and willingness to learn a bunch of unrelated stuff.

What was the reason you wanted to build a startup since you were young?

It’s the impact again. I’m the person who gets bored easily if I’m seeing that my job isn’t really changing much. I guess that was it. Also, the fact that startups are just incredibly hard to do automatically makes it interesting. Founding a startup is probably the most difficult thing you can do in CS, except for maybe some hardcore research. So yeah, I selected that to take on a challenge, I guess.

Why didn’t you go for typical big tech internships?

I had two reasons. One, I wanted to do high agency things with a big degree of responsibility, where it’s just a small team and everyone is responsible for the ultimate outcome of the project, not some smart person above you in the hierarchy always telling you what to do. Two, I really hate playing fair games, as in, applying in the general pool and going through the same hoops as everyone else. I prefer knowing someone who vouches for me to fast-track me in.

By that logic, I got into exciting projects I might not have if I just applied normally. I got into TUM.ai, MIT, and YC this way. If you want to end up in big tech, standard internships are probably fine. But since my mission since I started seriously coding at 18 was to found a company, I wanted to do things with the most agency possible.

AI Research at MIT

After your bachelor’s, you didn’t go for a master’s degree. Why is that?

The only reason I graduated bachelor’s was that I needed a passport of a first world country, which is Germany. And one of the prerequisites for that is a degree. If I was born in the European Union, I wouldn’t have even done B.Sc. I would have definitely dropped out.

How did you end up doing your bachelor thesis at MIT?

Again, I didn’t play a fair game at all. MIT has like a 4% acceptance rate, but for me it was 100%. There was exactly one person applying for that research spot, and that person was me.

So the way I got into MIT CCI (Center for Collective Intelligence) was the two TUM.ai presidents that were presidents before me actually wrote their master’s thesis at CCI with the same professor — Peter Gloor. I’m really grateful to them for recommending me. And the reason it was a great fit is because CCI’s research objective is using AI for the intersection between psychology, management and AI. They’ve been working on this for the past couple decades. And so my AI experience, plus my management experience with TUM.ai made me an ideal candidate for this research lab.

What was your research on at MIT?

My research itself was a pretty unique application of AI, actually. It was on how emotional entanglements and interactions between people on a team influence the team’s productivity.

We had some unexpected findings. For example, if there’s like one rockstar on the team that does everything and pushes a lot and talks a lot and everyone else is kind of slacking off, that team is not going to perform well. However, if the work is distributed more evenly but there’s this one person on the team that gives other people some nudging and they feel some pain or fear or anger, but like just a tiny bit of it, that team is going to be the most productive. On the other hand, everyone on the team is just always happy and jolly all the time, that team is not going to perform well.

These insights were well-known to CCI from their 20+ years of research. However, I helped prove them numerically for the first time. They set up an experiment with a couple dozen teams from different universities consisting of people that didn’t know each other, asked them to play a specially developed computer game, and ran my software that used AI to gauge their emotions and interactions. The reason why we had people play a computer game is that it gave us objective results for every team, so we could compare the productivity numerically and draw those insights.

I might or might not turn this into a couple of research papers. Considering that I’m now in YC and working on my startup full time, I’m probably not going to have time to publish this.

What are your thoughts on research vs startups?

It takes a special kind of person to do research - you have to be willing to invest years into something that might bear no fruit. That’s probably not me. I appreciate how tight the feedback loops are in startups. Actually, one of the most satisfying aspects of CS for me is exactly this, how tight those loops are - my dopamine receptors fire every time I write a tiniest function and it works. So I would be a great fit for startups, which is what I’m doing.

Another issue with research is also that there’s this incentive misalignment everyone’s talking about in academia. It forces people to optimize for e.g. number of publications, not quality.

Building the Future of AI at One Interface

Why did you decide to work on One Interface and what makes you passionate about it?

It seems very logical that the next step for AI development should be AI that knows the context of your life — or at least of everything you’ve been doing on your computer. People are trying to do this in various forms, but I think what we’re doing is the correct approach for the capabilities that large language models offer.

The models just want to learn. But no one is gathering the context data from you and your life to teach them. We have the data collection nailed down by now, and are experimenting with the use cases we want to build to maximize the value proposition. Automatically extracting tasks from your communication with other people, helping you productively work

We’ll be doing that for both individual users and teams. One of the good use cases for One Interface, in fact, is that you can share some of your context with your teammates, effectively making a self-creating knowledge base that looks like Notion but updates itself in real time. You won’t have to explain, like, your reasoning behind why you made such and such decisions, or you won’t have to spend weeks onboarding a new person.

Also, my CTO Robert is a great co-founder. We’re both not perfect, but we work around our issues and make a world class team.

I’ll end this with a loosely related quote from Sam Altman, who said something along the lines of: the future of AGI or ASI is each person or each team fine-tuning it on their own data, so it becomes like an intern that watches over your shoulder, looks at everything you’re doing and then learns from that. So he thinks the future of AI is going to be highly customized AGI, and I share that opinion. That’s why we’re building it.

How do you feel about the potential of your startup failing?

I would definitely regret it because of all the time, effort and heart put into it. But I would do it again and again and again. Tenacity is the only sensible way to approach startups. I’m going to feel the loss anyways, but it’s something I want to pursue for my whole career, so I’m not planning on giving up if it fails.

What does your typical day look like now?

These days I mainly sleep, code and eat. I wake up at a random time, code for a random amount of time. Maybe take a meeting, maybe not. Then maybe eat and chill for an hour with my girlfriend, then code or sleep again. I definitely don’t operate on a 24-hour schedule.

I do those bursts of coding that could be like multiple hours of me just getting in there and concentrating. Then I sleep for a couple hours and recharge, then do it again.

Advice for Ambitious Students

Looking back, is there anything you would change or do differently?

I wish I focused on one thing, CS, earlier instead of doing music, then physics, then STEM. I started CS at 18, so I’m still not at the level of some really talented people, especially some I met at MIT, who started coding at 10. You literally can’t beat someone with 3 times as much experience, even if you have significantly higher intelligence.

There’s still this gap, which of course is going to get smaller as we age and progress in our careers, but it’s still there. I guess that’s what I would have done differently - focus on one thing and decide on it earlier. The Ukrainian high school system wasn’t very conducive to that though, I had like 19 subjects in my graduation certificate. But I also should’ve known I wanted to do it earlier.

What advice would you give to younger students?

If there’s one thing I’d like to share with people who are younger than me, it would definitely be don’t think that you have enough time to figure it out. And I don’t even necessarily mean AGI, even though that too is basically consensus in the Silicon Valley circles that we’ll have it in like 5-7 years tops.

Don’t feel you have enough time because, you know, life’s too short. When I was younger, I actually didn’t realize how long it took, how much consistent effort it took to get really good at a certain thing. Don’t underestimate that. Don’t think you have enough time to try 10 things and figure out which one you like. Pick one thing early, stick to it, get good at it.

There’s this golden decade which starts when you graduate college with basically no commitments and ends when you have a family and kids. That decade is when you can achieve outsized impact. It’s arguably the most important time. If I had a child, I wouldn’t be half as risk tolerant as I am now. So enjoy it while you’re young.

---

Closing notes

Hey everyone, Arnie here :) Thanks for reading this week’s blog (and for the ETHZ students, good luck with exams!)

As always, be sure to vote what you thought of this blog: