Debapriya-source/nyaya-gpt

Combines ReAct reasoning with RAG retrieval over a FAISS vector database of the Indian Constitution and BNS, allowing the agent to iteratively refine queries across multiple thought-action cycles before generating answers. Implements multi-step legal document reasoning using LangChain's create_react_agent with Groq's Llama 3 LLM, embedding documents with HuggingFace's all-mpnet-base-v2 model for semantic search. Chunks and indexes legal texts with RecursiveCharacterTextSplitter, then surfaces a Streamlit interface for conversational legal queries with built-in error handling and memory across conversation turns.

No commits in the last 6 months.