AI Agents Assist Scientists at Facilities

Modern scientific facilities deliver extraordinary capabilities through increasingly complex instruments. New users face steep learning curves navigating proposal systems, safety protocols and intricate experimental setups. This operational complexity makes it challenging for scientists to design experiments that effectively leverage advanced instruments. Researchers at X-ray light sources, nanoscience centers and leadership computing facilities need better ways to access the expertise embedded in facility documentation and operating procedures. 

CALMS addresses this challenge by combining large language models with facility documentation and instrument control systems. The AI retrieves relevant information from technical documents to answer questions accurately. Without this documentation context, AI models hallucinate answers or decline to respond. With context, CALMS achieved relevant, truthful responses on 14 out of 14 test questions about experimental planning at facilities including the Advanced Photon Source, Center for Nanoscale Materials, and Argonne Leadership Computing Facility. The system maintains conversation history, allowing natural follow-up questions that build on previous exchanges. 

Beyond answering questions, CALMS takes action through instrument control. Researchers demonstrated this on an actual X-ray diffractometer at the Advanced Photon Source. When a user requested specific instrument positioning, CALMS queried the Materials Project database, calculated the required motor positions, and moved the diffractometer tasks that normally require manually looking up data and entering multiple commands. This demonstrates how conversational interfaces can streamline experimental workflows while keeping scientists in direct control.