Tommaso Dreossi
I build AI for scientific problems.
At Arena Physica, I'm developing a foundation model for electromagnetism, surrogate models for EM simulation, and inverse design pipelines using generative diffusion models.
Previously at insitro, I led ML for ALS target discovery and built self-supervised and computer vision models for high-content cell imaging and omics.
I did my PhD on reachability analysis for nonlinear dynamical systems and spent three years at UC Berkeley developing testing techniques for AI-enabled robotic and cyber-physical systems.
Writing
- A Control-Theory Taxonomy of Physical AI Jul 2026
Mapping the hyped Physical AI space onto the control loop.
- Gradient Descent into Confusion Mar 2026
What backpropagation taught me about how fields name things.
- The good, the bad, and the ugly of data science and ML in tech/biotech/academia Aug 2024
A comparison of career experiences for computational scientists across big tech, biotech, and academia.