I built JURIS‑IQ, a production‑grade AI system for legal contract analysis that started and was last updated in March 2026. The code lives in a GitHub repository owned by me (@sarthaksingh17) and is written in Python.
The repository is a repo on @sarthaksingh17's GitHub.
JURIS‑IQ deploys 5 AI agents that analyze it from every angle — legal risk, financial exposure, regulatory compliance, and precedent search — before a Judge agent synthesizes a final verdict with a confidence score.
Round 3 — agents defend or revise with new evidence.
Agent orchestration | LangGraph (typed state machine).
LLM | LLaMA 3.3 70B via Groq API.
Legal NLP | InLegalBERT (trained on 5.4M Indian court docs).
Vector search | FAISS (IndexFlatIP).
Metadata store | DuckDB (embedded, no server).
API | FastAPI + SSE streaming.
Frontend | Streamlit.
Async | asyncio.gather() for parallel agent execution.
Legal corpus | CUAD dataset — 510 real commercial contracts.
Overall risk score (0–100).
Legal, financial, and compliance breakdowns.
Top issues with recommendations.
Final recommendation: APPROVE / APPROVE WITH CONDITIONS / REJECT.
Full 9‑argument adversarial debate transcript.
SHA‑256 tamper‑proof audit hash.