I'm an ML engineer with a research background, currently finishing my M.Sc. in Computer Science at TU Berlin and moving toward AI alignment and safety research.
I've spent some time building and deploying machine learning systems at the frontier of biology and materials science — most recently working on protein engineering with generative models like RFdiffusion, AlphaFold, and ESM. Along the way I've worked on transfer learning for soft robotics, adversarial ML and security, and generative diffusion models.
These days I'm increasingly drawn to a different set of questions. I think AI safety is one of the most impactful things to work on right now, and I find it genuinely fascinating to try to understand these models — what they're capable of, how they fail, and what their behavior tells us about intelligence itself. I'm especially interested in empirical research and mechanistic interpretability as ways to get real traction on these questions.
This page is a work in progress, and so am I — more projects and writing are on the way as I move deeper into the field.
- 🧭 AI alignment & safety — understanding and mitigating the unintended consequences of increasingly capable models
- 🔬 Mechanistic interpretability — reverse-engineering what models learn and why they behave the way they do
- 🧪 Empirical research — getting traction on hard questions through careful experiments and honest evaluation
- 🧬 ML for science — generative models for protein and molecular design
- transferable-acoustic-sensing — research project on cross-device generalization in acoustic sensing models for soft robotic actuators, framed as a domain-adaptation problem
- I also work on a creative research project exploring how writing and object theater can make AI alignment failures — emergent misalignment, harmful outputs, bias — physically and emotionally tangible
Most my work lives in private or institutional repositories. A few examples:
- Protein engineering at Cambrium — internal ML tooling and exploratory research adapting generative protein models (RFdiffusion, AlphaFold2, ProteinMPNN, ESM) for targeted protein design
- RFdiffusion conditioning — bachelor's thesis work introducing conditioning mechanisms into a generative diffusion model for proteins, presented as a poster at the European Rosetta Conference 2023
- Adversarial ML & LLM security — empirical investigation across vulnerability detection and adversarial robustness tasks (BIFOLD MLSEC)
- LinkedIn link
- LessWrong/AI Alignment Forum nonergodic
Happy to discuss any of these in more detail — just reach out.