SentinelRAG

SentinelRAG is a robust retrieval-augmented generation (RAG) system engineered to prevent sensitive information leakage when querying internal code repositories and documentation. It integrates hybrid retrieval (BM25 + FAISS) for comprehensive context, a two-stage policy engine to classify and manage query risk, and confidence-aware answer routing to prevent hallucinations. By combining these layers, SentinelRAG safeguards against adversarial or careless queries that could otherwise expose API keys, credentials, or other secrets, setting a new standard for secure, policy-aware RAG applications.

主な機能

ユースケース