| Abstract Scope |
The accelerating adoption of automation in materials science has highlighted the need for intelligent characterization systems. We introduce a Multi-Agent AI controller for an X-ray microscope (XRM) that integrates instrument operation with Python-based image analysis, a deterministic geometry solver, and a retrieval-augmented generation (RAG) pipeline. At its core, a large language model (LLM) orchestrates the system, performing complex user tasks such as selecting acquisition parameters, analyzing data, and managing experimental workflows. We evaluate system performance across common XRM use cases, emphasizing robustness and adaptability. Finally, we demonstrate integration of the LLM-controlled microscope into a Bayesian active learning loop for autonomous materials discovery. In this manufacture-characterize-update cycle, the LLM agent autonomously controls the microscope, interprets results, updates an active learning regression model, and selects the next candidate for synthesis and characterization. The generality of Multi-Agent LLM platforms is then discussed for expanding to other characterization methods and materials development pipelines. |