Empowering x-ray science with LLMs and agentic workflow

Foundation models (e.g., LLMs, VLMs) are revolutionizi X-ray science by enabling intuitive human–computer interfaces, fuzzy logic-based automation, and powerful data insights. Deploying them effectively, however, requires frameworks that integrate advanced AI with existing expertise. Nodeology addresses this challenge through a modular, graph-based architecture that merges AI-driven techniques with established methods, while maintaining crucial human oversight. Workflows can be shared, adapted, and versioned using Nodeology’s template system, fostering collaboration and reproducibility. We have used Nodeology at the Advanced Photon Source to automate ptychography and X-ray fluorescence (XRF). In ptychography, AI-based workflows reduce trial-and-error by recommending reconstruction parameters, generating code, and analyzing results iteratively. Similarly, our XRF copilot program can control instrumentation for real-time optimization. Moreover, these AI-centric workflows can even facilitate autonomous research, from literature insights to exploration of novel algorithms.