Speaker
Description
We present the first application of AI agents to the design and optimization of detectors for high-energy physics experiments. Our bi-level optimization framework vertically integrates detector geometry, front-end digitization, and high-level reconstruction parameters within a differentiable, end-to-end simulation. Using a segmented dual-readout crystal electromagnetic calorimeter as a case study, we investigate the ability of large-language-model-driven agents to manage and guide the optimization workflow. The agent analyzes previously evaluated configurations, identifies performance trends, and proposes new trials across the full detector-design parameter space, extending beyond conventional Bayesian optimization.
We find that current frontier reasoning models, even without providing expert experiment-specific knowledge, can execute complex simulation workflows and proactively identify relevant directions for further investigation and design improvement. We demonstrate that an AI agent can identify favorable detector configurations under competing physics-performance objectives, improving key physics metrics while potentially reducing scientific labor, computational expenditure, and detector-development costs. This study establishes a foundation for increasingly autonomous detector optimization and represents a step toward the first fully AI-designed detector for a future scientific facility.
