Automata for Physics
I explore how LLMs can accelerate physics research by connecting knowledge across fields with analytical calculations, numerical simulations, and experiments. The aim is to develop agents that formulate useful questions, carry out investigations, and revise their approaches through verification and feedback. I am interested in both collaboration with human researchers and autonomous workflows, including how to evaluate their scientific reliability.
Papers
-
Evaluating Verified Autonomy in Quantum Engineering
Preprint · arXiv:2609.17439 (2026)
We introduce Quantum-Harbor, a virtual laboratory that evaluates agents through their interactions with quantum systems, checking both the operations they perform and the conclusions they report. QIQCBench supplies 49 expert-designed tasks covering calibration and control, error correction and compilation, sensing, and networking. Tests of 17 agentic systems reveal substantial variation in verified performance and show that reliable autonomous operation remains an open challenge.