Speakers
Description
We present Dr.Sai, a novel multi-agent system powered by large language models (LLMs), designed to automate full-chain physics analysis at the BESIII experiment. The system directly interprets a physicist's natural language query (e.g., "Measure the J/ψ mass spectrum and branching ratio"), autonomously decomposes it into structured subtasks (data skimming, fitting, etc.), and orchestrates the generation and execution of corresponding code. Specialized agents manage workflow planning, scientific code generation, job submission to computing clusters, and result validation, creating a closed-loop, reproducible analysis pipeline.
Benchmark tests on established BESIII processes, such as J/ψ decays, demonstrate the system's practical reliability. Dr.Sai successfully completed multiple, distinct end-to-end analyses, achieving an overall workflow success rate exceeding 90%. This work provides a concrete, scalable blueprint for "AI scientists" in experimental HEP. It showcases a transformative shift from AI as a tool to an autonomous partner, significantly accelerating analysis cycles and enabling scalable exploration of complex data. We discuss the integration of this agentic framework within IHEP's computing ecosystem and its potential to redefine collaborative human-AI discovery in particle physics.
| I read the instructions above | Yes |
|---|