A bit more about me

I’m David Podolskyi, an engineer based in Munich. I studied Information Systems and Mathematics at TUM, and I’ve worked on software and machine learning since 2020.

I like being new to a subject: reading enough to understand the problem, trying things, and getting something working. That’s taken me into fairly different areas, including brain imaging, semiconductor design, and game-playing agents.

What I’ve been doing

I’m now a software & ML engineer at Apple. I joined as an intern, built a production system for turning software issues into code changes, and stayed on full-time. I also train models to handle parts of that process.

Before that, I wrote my Bachelor’s thesis at Harvard Medical School. I researched how neural rendering could help align a surgeon’s camera view with brain scans. It was a chance to work on a problem where the engineering and the science were closely connected.

At Infineon, I worked on deep learning for chip design and quality assurance. At Expleo, I built data pipelines for BMW’s autonomous-driving work, processing more than a million data points a day. I’ve also built language-model products for legal and investment workflows and worked with startups at UnternehmerTUM.

The TUM.ai years

I joined TUM.ai in 2020 and was president from 2022 to 2024. I helped grow it from a small group of students into a community of 300+ members. We ran summits and hackathons, coached startups, worked with industry partners, and started a research department.

With KNUST in Ghana, I helped organise and teach AI workshops. I also co-organised a defense hackathon with more than 400 participants and helped build a hotline for Ukrainian refugees with volunteers at AppliedAI.

I still like hackathons as a way to try an unfamiliar idea with other people. Three first-place finishes so far, and a top 5% finish on Kaggle with a Pokémon agent. At EDTH Paris in December 2024, our team won second place with a magnetic and inertial navigation prototype. I led implementation and presented the indoor demonstration.

What I want to understand next

I want to understand what models do internally, how that changes during training, and whether that understanding can help us build better ones. I’d like to work on explainable AI, mechanistic interpretability, and alignment, and explore how these connect to models trained on biological data.

I also want to build a company and keep doing work that helps Ukraine. I’m still working out what that looks like.

There’s also pickleball. And, hopefully soon, more padel 👀