About me

I'm Raphael, a machine learning researcher from Germany, currently based in Cambridge, Massachusetts. My research at MIT focuses on reinforcement learning and multi-agent systems, alongside research collaborations at Harvard.

What draws me to machine learning

The questions that draw me in are often mathematical: how structure shapes what a model can learn, how uncertainty affects decisions, and why learned behavior generalizes. I'm especially interested in machine learning theory, statistics, and probability theory.

During a research internship at Porsche, I worked on multimodal learning and the post-training and fine-tuning of transformer and diffusion models, including neural surrogates for physical simulation. I also contribute to a Harvard research project on LLM interpretability and language bias, supervised by Ryan Badman.

I'm also interested in automated research: how agents can propose hypotheses, run experiments, and evaluate their own work.

Building alongside research

I co-founded PrognoseWerk to build forecasting software and work on Nyra, a personal AI project. I like working with small teams and seeing how research ideas behave in use.

Outside the work

I enjoy playing chess, running, and going to the gym. I also raced karts nationally and internationally, including the ROK Cup world final in 2017 and the Winter Cup in Italy in 2020. At the Winter Cup, I raced against Kimi Antonelli, now a Formula 1 driver. I still follow the sport.

I enjoy traveling and meeting people whose work is quite different from mine. The IDEA League Challenge Programme and volunteering with IT4Kids have been two ways of doing that.

I'm always happy to hear about an interesting research question, a startup idea, or a place worth exploring.