Document Extractor

A production-grade utility for extracting structured data from PDF and DOCX documents. Built-in hallucination verification, structural derivation, and multi-format support. Designed to drop into a production environment in under five minutes.

For CLI tools, document batch processors, or internal data science pipelines.

pip install pymupdf python-docx anthropic msgspec spacy
python -m spacy download en_core_web_sm

from contract_extractor import ContractExtractor

extractor = ContractExtractor()

with open("contract.pdf", "rb") as f:
    result = extractor.extract_from_file(f.read(), extension="pdf")

print(result["metadata"]["knowledge_graph"])

Deploy as a stateless extraction microservice. A simple_api.py is included for rapid rollout.

python simple_api.py

curl -X POST -F "file=@legal_agreement.docx" http://localhost:5000/analyze

Use the knowledge graph and verification logic to build custom dashboards: risk dashboards triggered by null-derived risks, visual timelines mapping obligations to a Gantt chart, or clickable audit viewers using reference_snippet and bbox coordinates.

Bias detection — Identifies clause bias (buyer/seller favorability) using field facets.

Verification logic — Python-level grounding verification for every extracted quote. Hallucination guard for all outputs.

Object derivation engine — Automatically converts flat keys into rights, obligations, and risks.

Multi-format structural intelligence — Treats DOCX headings as virtual pages for consistent downstream processing.

1. Add your ANTHROPIC_API_KEY to a .env file.
2. Python 3.9+ recommended.
3. Documents are processed via Anthropic's API. Ensure your use case complies with your data sovereignty requirements.

Licensed under the GNU Affero General Public License v3.0.

This core engine is open source to drive adoption of verified document intelligence. Making the grounding and derivation logic public is an effort to eliminate black-box extraction in document processing.

qleric.com