Workshops Archive

LightshowAI-Agent: An MCP-Based Agentic Framework for Real-Time and Reproducible X-Ray Absorption Spectral Analysis at Scientific User Facilities


Workshop: AI4S: 8th Workshop on Artificial Intelligence and Machine Learning for Scientific Applications

Authors: Ozgur O. Kilic and Sairam Sri Vatsavai (Brookhaven National Laboratory); Huihuo Zheng (Argonne National Laboratory); and Anton Yanaky, Xiaohui Qu, Christopher Kelly, Chuntian Cao, Kari Barry, Daniel Allan, Lu Ma, Stuart Wilkins, Sara E. Mason, Meifeng Lin Meifeng Lin, Adolfy Hoisie, and Deyu Lu (Brookhaven National Laboratory)

Abstract: Identifying an unknown phase from X-ray absorption spectroscopy (XAS) data requires a multi-step pipeline spanning candidate retrieval, surrogate inference, spectral alignment, expert-informed ranking, and provenance tracking across independent databases, models, and services that scientists coordinate manually. We present LightshowAI-Agent, a framework that automates and orchestrates this pipeline. Scientific capabilities are exposed as typed tools through the Model Context Protocol (MCP), providing access to the Materials Project database, XAS prediction surrogates, and American Science Cloud (AmSC) model services. Versioned skill files carry domain-specific analysis procedures that experts can inspect and revise independently of prompts. Reproducibility is supported by logging agent and model activity through MLflow and analysis metadata through Tiled. We demonstrate a workflow identifying a TiO2 polymorph from Ti K-edge spectra streamed from the NSLS-II Quick X-ray Absorption and Scattering (QAS) beamline.


Back to AI4S: 8th Workshop on Artificial Intelligence and Machine Learning for Scientific Applications Archive Listing

Back to Full Workshop Archive Listing