PhD student · ETH Zürich

Making useful models.

I am a PhD student at ETH Zürich’s DISCO Lab, interested in making machine learning more useful in practice—by improving what models can learn, how well they generalize, and how efficiently they can do it.

I study how training data, architecture, and training methods shape model behaviour, including how LLMs can learn better from verifiable and semi-verifiable feedback. At the same time, I build efficient models—such as logic gate networks—and improve the code and algorithms around them by finding bottlenecks and making better use of available hardware. The goal is the same in both cases: models that deliver more capability for the compute, latency, throughput, and energy they use.