Featured
Active
Clause Lens
Insurance Policy Clause Extractor
Problem
Insurance policies are dense PDFs — extracting coverage facts, exclusions, and limits manually is slow, inconsistent, and error-prone.
What I Built
Full pipeline: PDF ingest, clause chunking & classification, concurrent Claude JSON extraction, embeddings, pgvector cosine search, labeled eval benchmark, and a built-in web UI.
Tech Stack
TypeScript
Node.js
Claude API
PostgreSQL
pgvector
Python
Express
Docker
LLM
RAG
Outcome
End-to-end policy analyzer with retrieval accuracy targets above 80% and field-level extraction scoring on a labeled benchmark set.
Results
- PDF → chunk → extract → embed pipeline
- Semantic search with clause-type filters
- Labeled evaluation report