JurisRAG---Citation-Enforced-Legal-RAG-System

JurisRAG is a citation-enforced legal Retrieval-Augmented Generation (RAG) system designed to produce grounded answers from Indian legal documents. It is a production-grade deployable RAG system that combines structure-aware document ingestion, hybrid retrieval, citation enforcement, grounded generation, evaluation, and containerized deployment into a single end-to-end system. Every generated answer is grounded in retrieved legal evidence and required to provide citations to the underlying statutory sections. The system also evaluates its generated answers across important RAG quality dimensions including relevance, correctness, groundedness, completeness, citation correctness, citation completeness, context utilization, abstention quality, and hallucination. JurisRAG uses Azure AI Search for vector retrieval, Azure OpenAI embeddings (text-embedding-3-large with 1536 dimensions), and GPT-based grounded generation with the gpt-6-luna model. It employs hybrid legal retrieval combining traditional lexical/BM25 retrieval with vector similarity search, and uses Reciprocal Rank Fusion (RRF) to combine retrieval signals. The system includes structure-aware legal chunking that preserves act, chapter, section, subsection, marginal note, and content-type information, and can explicitly indicate when retrieved evidence is insufficient instead of fabricating an answer. Evaluation-driven development was used, with retrieval and answer-generation stages evaluated independently, and a 50-question golden dataset for answer-level evaluation. JurisRAG is distributed as two Docker images (backend and frontend) and includes GitHub Actions CI/CD for container builds and publication.