About me

I've always been drawn to the question of how machines can learn to reason about the world. That question has shaped my path from theoretical physics of complex systems into machine learning and artificial intelligence, and today I lead AI Research at Axiomatic AI, where my team builds agentic systems that write and verify formal proofs in Lean4.

My research started at the boundary of physics and artificial intelligence. I studied theoretical physics at Sapienza University, focusing on the statistical mechanics of spin glasses. Subsequently, during my PhD at the Max Planck Institute, I developed machine learning methods for artificial scientific discovery: extracting meaningful features from raw data, designing experiments autonomously, and using program synthesis to decompose quantum unitary matrices. At the Flatiron Institute in New York I built on these ideas at astronomical scale, training multimodal transformer models that learn shared representations across images, spectra, and time series.

I've also worked on reinforcement learning at Google DeepMind, studying how agents navigate environments where rewards are extremely sparse. These are agents that learn to explore by seeking novelty, trying things they have never tried, and by being curious, predicting what will happen when they do, much like a scientist would.

In my work, I like moving fast from idea to working system, building end to end, and getting the details right without losing sight of what the system is actually for.