Artificial Intelligence and Autonomous Scientific Reasoning

My work in artificial intelligence focuses on the development of agentic systems for scientific reasoning, with particular emphasis on simulation-native autonomy in experimental physics. Rather than treating AI systems as tools for workflow automation or code execution, I investigate how large language models can participate directly in hypothesis formation, experimental design, simulation-driven evaluation, and explanatory synthesis under physical constraints.

This research program is instantiated in GRACE (Generative Reasoning Agentic Control Environment), an autonomous AI framework designed to operate at the upstream level of experimental design in high-energy and nuclear physics. GRACE treats detector design as a constrained search problem under physical law, using first-principles Monte Carlo simulation as an epistemic oracle rather than relying on predefined optimization heuristics.

The broader aim of this work is to understand how AI systems can construct, evaluate, and revise explanations in domains where correctness is determined not by benchmarks or labels, but by consistency with physical theory, statistical structure, and simulation-grounded evidence. I am also involved in some fronts in medical use of machine learning models.

GRACE Whitepaper (ArXiV)
GRACE Project Compute Grant
Oncology ML Project Compute Grant