NITK's rhetorical-role RAG boosts legal QA on Indian Supreme Court judgments
New RAG framework splits legal texts into rhetorical roles for smarter retrieval and answers.
Deep Dive
Key Points
- Uses rhetorical-role chunking, fusion retrieval, and cross-encoder reranking to boost legal passage retrieval
- Supports multi-turn conversations via chat history, query classification, and query rewriting
- DeepEval evaluation shows strong contextual recall and answer relevancy for explainable legal AI
Why It Matters
More reliable legal AI could help lawyers and citizens quickly navigate complex Indian Supreme Court precedents.