The Evolution of Video Editing and Personal Gaming Experiences

Currently: Co-founder (CTO & CRO) @ Mimic Robotics Studies: PhD Computer Science @ ETH Zürich (Soft Robotics Lab, ETH AI Center Fellow) ‘21-’25, MSc Data Science @ ETH Zurich ‘18-’20, BSc Computer Science @ University of Milan ‘15-’18 Experiences: CTO/CRO @ Mimic Robotics, Research Intern @ Oracle ‘20 Origin: Bergamo, Italy Links: LinkedIn, Twitter

Mimic Robotics is hiring! They have an exceptional team of engineers and working on cutting-edge robotics. Consider applying!

Supported by Founderful Campus! They run the VC Academy program, selecting top students from Swiss universities that invest in the hottest university startups.

Growing up in Bergamo & Making Short Films

What were you doing in high school?

I went to a scientific high school in Bergamo, North Italy. During that time, I had this random fixation on wanting to do visual effects movies. I had been doing home movies since I was a kid, and I really wanted to do high-budget visual effects movies. I discovered After Effects because of Freddie Wong (now RocketJump) - I was really obsessed with his channel.

Between my third and fourth year of high school, when I was around 16-17, I spent all my time making these videos. I was the only one pushing this forward - all my friends didn't care much, they just showed up, did their horrible acting, and I recorded everything and did all the visual effects. The last video I made was about Crysis, the video game. It was a 10-minute video that took about three months to make, which seemed like an absurd amount of time back then.

How complex was the video editing you were doing?

Considering this was done around 2012-2013, the software wasn't that good. For example, if you wanted bullet holes on walls, you had to install plugins for After Effects with motion tracking. You had to pick a spot on the wall, select the width for the motion tracking algorithm to follow, and it would mess up half the keyframes - then you'd have to change it manually to make it fit. Now you probably have some deep learning based tool in Adobe that just does it for you, but back then it was much harder.

You were also big into gaming - how did that influence you?

I had what I considered an absurd amount of hours in Counter-Strike - around 800 hours (though I know people who did thousands). I think video games are good when you're in high school because you don't really have a lot of exciting things going on. When you play video games, you actually get to do something cool and optimize things - like building a super optimized factory in Factorio. You can have an objective and actually do something about it when you don't have any money and are living with your parents.

That said, I played way too much even during my bachelor's studies, which definitely shouldn't have happened. Counter-Strike specifically is not very creative - you're doing the same stuff over and over to get good at it. I remember one day going to my friend saying "I will never play this game ever again, this is horrible, it just makes me rage." From that day, I just stopped.

Studying Computer Science & AI

How did you decide to study computer science?

Interestingly, I knew I wanted to do computer science when I was around 8 years old, even though none of my parents did more than high school. I just liked computers and had been playing with them since I was a kid. In high school, I was reading lots of books about AI and what AI would be - even before deep learning was really a thing. I really got into AI, so it came very naturally to do computer science for my bachelor's and then focus on data science at ETH.

What was studying at University of Milan like?

Throughout my bachelor's I was commuting one and a half hours each way, which I absolutely do not recommend if you can afford not to. You miss out on many activities outside of studying. At UniMi, you get to do lots of theory - learning everything from scratch about computation theory, formal systems, and programming languages. You don't really get to do lots of activities like startup clubs - that's more common at Politecnico which is considered the tech hub of Italy.

I think if you want to actually grow as a person doing computer science and AI and you’re not from one of the big hubs, you have to basically make it all up by yourself. We had projects but sometimes they were quite outdated - I remember there was an artificial intelligence course using Prolog from the 80s. These were more like coding exercises rather than building something real.

How did you decide to apply for a master's at ETH Zurich?

I was doing some AI courses on the side through platforms like EDX. I had a list of schools I thought were cool - there was a CDT program at University of Edinburgh that combined Masters and PhD, ETH Data Science, and UCL. Nobody was telling me that I should do this or that. When I talked to people about doing a Master’s outside of Milan, they were like "oh, that's interesting, I never thought about that." Nobody even conceptualized that you can go somewhere else.

I think it's very important that you figure this stuff out on your own and apply to many places. I basically was just randomly stumbling through it. I did a ridiculously low number of applications - I applied to three things and got into one. Even for ETH, I only applied to data science. I should have applied to computer science as well. There was no culture of like, I don't know, the US-type culture where you have to get into Ivy League. You have to make it up yourself.

Masters at ETH Zurich

What are some of your best memories from your time at ETH?

I loved going to hackathons during the Masters. I didn't do too many, and that's something I regret - I should have done more. We were winning these hackathons like VIS Hack, PolyHack, ... very often. It was very fun. I think this is very good because when you're young enough that your body allows you to do all-nighters, you should do them.

To me, that was one of the best things - actually just doing these random apps in hackathons where you learn how to do some project that is not just a school project. Because even in the ETH Masters, all the core stuff is still focused on more school-type projects. Even though sometimes in the deep learning course, they're like "oh, if you're really good, it should become a paper." But it never has this energy of "you want to make something, you have to deliver it in the next three days."

Internship at Oracle & Starting a PhD

How was your experience at Oracle and how did it influence your decision to do a PhD?

The internship was complicated because COVID happened and the very best positions I wanted got messed up. I just wanted to get an internship done before doing the PhD because I wanted to get some experience in a more corporate place. Not because I wanted to work in corporate, just because I had never done anything that remotely resembles a real "grown-up job."

I think there was a different culture during COVID where everything was remote. As a very junior person, the remote thing doesn't work - you need somebody to talk to for any random stuff coming up. If you have to do a call for each of these things, it's very awkward. For juniors, remote work doesn't work. You actually need to learn, and then you're completely stuck.

But regarding wanting to do the PhD or a startup - I may be a bit of an unusual case compared to your model podcast guest. I was always ambitious in some regards: I want to do X, I want to do videos, I do them even if nobody around wants to do them. I want to do computer science, I will do it. I want to go abroad, I will do it. But I never really had the startup push at the beginning. I basically just wanted to do research in AI and work on AGI. At the time, the path was clear - do a PhD, try to do as much as I can, and try to work somewhere that actually does this seriously, like Google where they have the resources to make it work.

Founding Mimic

What does your startup Mimic do?

We think that with current advances in generative models for robotics, you can have general purpose robots that can solve various degrees of manual labor that currently props up the world economy. Manual labor is usually not a dream job - people don't necessarily want to do most manual labor. There's going to be problems in advanced economies with labor shortages and aging demographics.

If you want to maintain and increase the standard of living of today and have abundance, you will need to somehow automate generalizable manual labor. Rich countries can get away with it right now by exporting it to other countries, but ultimately as the entire world gets rich, nobody will want to do this.

We combine expertise in building dexterous humanoid hands with training models using imitation learning. The idea is that just with imitation learning, you can already approximate to high success rates enough tasks that you can have a successful startup with customers today. You take the hand that is designed to mimic humans and you can then train it to actually mimic humans using data.

How did you transition from PhD to founding a startup?

I decided to do the PhD because I thought it was the best path to enter big tech research and to meaningfully work on AGI. But then, everything started happening around ChatGPT and LLMs and agents. I suddenly realized that working in the space became ‘easy’ in a way that you could actually drive something individually.

I also realized that if you actually want to be at the forefront, even in terms of research & building transformative AI that could change the world, papers & “acedemic-minded” research was no longer the best way. It became much easier to raise venture capital money and have a team that gives you the right resources to solve the problems you want to solve.

How did you meet your co-founders

I met Stefan and Stephan through the Soft Robotics Lab. They were both already in the startup scene and very strong, they had this whole idea focused on hands and hardware. I wanted to do startups too, and had my own ideas about how to do end-to-end models for manipulation. We met at exactly the right time to make it work and raised the pre-seed quite early on.

What are some key challenges you’re facing in the startup?

I can talk about some examples very specific to what we do. For example, the whole data collection topic - how do you make sure tasks are being collected correctly and then evaluated? How do you run evals for the robot AI? If evals are hard for LLMs, they are absurdly hard for robotics.

When you submit a robotics paper to a conference, there are no benchmarks that make sense. Your robot model is always going to do better than some other models applied to your robot because the robot is in a different setting - different video, lighting conditions, everything. Even the robot morphology is usually different. Setting up evals within the organization somewhat autonomously without micromanaging is very hard.

Building and Managing a Deep Tech Startup

How do you approach delegating tasks as a founder?

I think delegating is very hard, especially coming from a PhD background.

Firstly - when you want to find people who are really good at what you're good at, it's really challenging. You have to be really careful with hiring and finding exactly the right person. It takes forever and feels frustrating because you'll end up throwing away work when people don't work out, or pursuing candidates who ultimately don't join. But if you don't invest this time in hiring, the outcomes are worse overall.

For delegating itself, you need to quickly establish feedback loops to understand what each person is good at. If you let things go for too long without checking in, you never know what will come out on the other side. We're still quite young as a startup so I'm still figuring out delegation, especially since we're working on novel technical challenges where there aren't established management frameworks.

How do you ensure everyone shares the same vision?

You can't run the startup like a research lab where PhDs have their own agenda and need first-author papers. The only way it works is if there's a common vision - if you don't share that vision, you're not in the startup. That's the only way.

The startup needs to work on one thing and do it really well. It can be research-flavored, like solving fundamental technical challenges, but you need to have a thesis of how you're going to solve your problem and stay focused on that. If somebody wants to try random methods that don't align with the core mission, you can't accommodate that. You have to ensure alignment from the beginning and build your team based on that vision.

What does Mimic look for when hiring talent?

Our needs are rapidly changing, but I think there's a lot of value in looking beyond just credentials and seniority. What we're doing is so new that there isn't really anybody who has been doing this for 5 years. The people who have been doing what we do for years like us are probably at companies like Physical Intelligence raising massive amounts of money.

We look for people who are strong technically and excited about what we want to do, making sure they fit with our vision. In terms of actual roles, we hire software engineers, AI engineers, mechanical engineers, and electrical engineers across the board. Being from ETH and within our network helps because we already know the person, but overall, if you share our vision for robotics and you're really good, then it's a good fit. We're trying to get people who are exceptional at what they do.

The cultural alignment is crucial - you need to be in the office, especially as a startup under 50 people. Remote work doesn't make sense at our stage. You need to be there, work hard, learn and grow and get stuff done, while also having time to think strategically and recharge.

Advice for Students

How can CS students get into Robotics?

You should unironically just do a course on ROS (Robot Operating System) basics. Then buy one of these low-cost robot arms to do some experiments. Learn about some basic controls and ROS. You should already know about LLMs and vision language models and generative diffusion - that's very important. Then there are papers on applying this stuff to robotics, like end-to-end robot learning. But honestly there's not that much in-depth theory. It's just about getting some experience in robotics and getting started with it.

What's your advice for ambitious people looking to start something?

I think if you want to do a startup in robotics, it's a good time for it, especially if you're in Europe where there's barely any competition. You should never think the classic efficient market hypothesis fallacy - "oh if it was so easy somebody would have done it already." That's not true. If you think you have an insight into something, most likely other people haven't figured it out - you should just do it.

It's easy to assume that because your entire life there were grownups, then people have figured out stuff. It's not true. Just do what you need to do. You don't have lots of competition currently, at least unless you're exactly in SF where everybody's always doing the latest things. But even there, I wouldn't overestimate SF compared to here as an ecosystem. It’s just harder to compete for local talent there, but even there there’s a shortage of people with the right ideas.

Editor notes:

Hey - Arnie here! Hope you enjoyed this episode :))

As always, lmk what you think of this blog + podcast, & any future guests I should interview or questions I should ask.

Running this takes a huge amount of effort with my busy schedule, but I find it incredibly valuable to push the ecosystem forwards & show anyone can be ambitious & build great things.

Ambitious x Driven enables ambitious EPFL & ETHZ students to learn from students/alumni who are a few steps ahead.

Show some <3 by sharing this blog with ambitious friends!

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.ambitiousxdriven.com