Indian Legal AI Platform — Hybrid RAG + LangGraph
End-to-end AI legal research and drafting, grounded in statutes and Supreme Court case law
An end-to-end AI legal research and drafting platform for Indian law. A Hybrid RAG retrieval layer (Voyage law-2 embeddings + Qdrant + BM25 + an LLM reranker) runs over 362 curated chunks spanning 41 statutes, 50 Supreme Court cases, and a 30-edge citator graph. A LangGraph agent pipeline classifies, clarifies, retrieves, synthesizes, and verifies every answer, with a citation verifier that strips hallucinated cites before they ever reach the user. The same grounding powers drafting: 21 contract types and 5 legal notices with PDF generation and a clause-level, RAG-grounded audit and edit-improve loop.
The pain this had to solve
Generic LLMs are dangerous for legal work in India: they confidently invent case citations, paraphrase statutes that do not say what they claim, and have no idea whether a precedent has been overruled. A wrong cite in a brief or contract is not a cosmetic bug — it is a liability.
The goal was a system that researches and drafts against real Indian statutes and Supreme Court case law, refuses to fabricate authority, and produces work product a lawyer can actually file — research memos, contracts, and notices — without manually re-checking every reference.
What I built — the architecture
Voyage law-2 embeddings in Qdrant combined with BM25 keyword search and an LLM reranker, retrieving over 362 curated chunks across 41 statutes and 50 Supreme Court cases.
Every query flows through classify, clarify, retrieve, synthesize, and verify nodes, so ambiguous questions get clarified and answers are assembled and checked, not guessed.
A dedicated verification step strips hallucinated citations and runs a fuzzy quote-check against source text, so only authority that actually exists and actually says what is claimed survives.
A 30-edge citator graph links cases to the precedents they rely on or distinguish, giving the retrieval layer awareness of how authorities relate.
21 contract types and 5 legal notices generated with PDF output, each backed by a clause-level, RAG-grounded audit and an edit-improve loop.
What it delivered
The platform researches and drafts against 41 statutes and 50 Supreme Court cases with a citation verifier that strips hallucinated authority before it reaches the user — turning a tool lawyers could not trust into one that produces filable research memos, contracts, and notices without re-checking every reference.
Want results like this for your team?
Tell me what you want to automate. On a free 30-minute call I’ll tell you straight whether it’s worth building, roughly what it costs, and how I’d approach it — no pitch, no obligation.
Book my free 30-min AI scoping call →