DocuMind

The problem

A legal-process outsourcing firm triaged 8,000 contracts/year into 14 matter types by hand - slow, inconsistent, and impossible to audit.

What we built

We built a RAG-based classifier (LangChain + a vector store of precedent clauses) that reads each contract, cites the specific clauses driving its classification, and routes low-confidence cases to a reviewer.

The outcome

Classification accuracy reached 95% on held-out cases, triage time fell 80%, and every decision now ships with a citation trail for audit.