Beacon Legal
Reviewed 12,000 contracts in a weekend
Built a RAG-based contract review pipeline that surfaces obligations, risks, and renewal clauses in seconds.
- Client
- Beacon Legal
- Industry
- LegalTech
- Duration
- 10 weeks
- Services
- AI / ML · Cloud & DevOps
The challenge
What was broken.
Beacon's paralegals manually reviewed each contract for obligations, indemnification clauses, and renewal terms — taking 45 minutes per document. Their backlog was 4,800 contracts and growing.
The solution
What we shipped.
- Designed a 3-stage pipeline: layout-aware chunking → embedding → structured extraction
- Used hybrid search (BM25 + semantic) over pgvector for citation-grade retrieval
- Built an evaluation harness with 200 hand-labeled contracts for regression testing
- Added a human-in-the-loop UI for ambiguous extractions
- Deployed on AWS Batch for cost-efficient bulk processing
PythonOpenAILangChainpgvectorFastAPIAWS
The result
Numbers from their analytics.
12,000
Contracts processed in 48h
94%
Extraction accuracy
45→2 min
Per-contract time
$280k
Annual labor saved
“Our paralegals went from drowning to leading the AI rollout firm-wide. That's the dream.”