HERMES Multi-Agent Framework for Large-Scale Geoscience Extraction
August 17, 2026
HERMES uses a coordinating LLM to extract structured data from ultra-long, complex scientific documents by integrating text, tables, and figures. The framework achieved 0.90 F1 for entities and 0.91 for attributes when processing the 55-volume Treatise on Invertebrate Paleontology.
HOW THIS AFFECTS YOU
●
builderYou can use this multi-agent approach to automate data extraction from massive, unstructured scientific corpora.
●
researcherThis demonstrates a method for maintaining high F1 scores during large-scale, document-level information extraction.