google/langextract

LangExtract solves the problem of extracting structured data from long, unstructured documents with high recall and traceability, overcoming the 'needle-in-a-haystack' challenge. Unlike generic LLM prompting, it enforces consistent output schemas via controlled generation and provides source grounding for every extraction, enabling easy verification and reducing hallucination risks.

Developers can build applications that extract entities, relationships, and attributes from clinical notes, radiology reports, legal documents, or any unstructured text. Real-world use cases include structuring patient medication lists, extracting character relationships from literature, and organizing radiology findings. The ceiling includes processing thousands of entities across long documents with interactive HTML visualization for review.