AgentCAT Extracts and Analyzes Catalytic Reaction Data

This arXiv paper (submitted Feb 10, 2026) presents AgentCAT, an LLM agent that extracts and analyzes catalytic reaction data from chemical engineering publications. It introduces a schema-governed extraction pipeline, a dependency-aware reaction-network knowledge graph, and natural-language querying and visualization, evaluated on roughly 800 peer-reviewed papers. The system aims to alleviate data bottlenecks and enable cross-paper mechanistic and outcome analysis for catalysis researchers.