A RAG-based AI Legal Assistant designed to provide tailored legal information from a private knowledge base. It features an adaptive user interface, a sophisticated retrieval-scoring-rewriting loop for accuracy, dynamic question suggestions, and a self-improving FAQ system.
Key Features
-
Interactive Chat Interface: A user-friendly chat application built with Streamlit.
-
Adaptive AI Persona: The assistant adjusts its communication style and response depth for Legal Professionals, Law Students, and the General Public.
-
Document Management Dashboard: An interface to upload PDF documents, which are then processed, chunked, and stored in a vector knowledge base.
-
Agentic RAG: The system uses a LangGraph-powered agent that:
- Retrieves relevant document chunks from a Pinecone vector store.
- Scores the relevance of the retrieved context against the user's query.
- Rewrites the query and re-retrieves if the initial results are not relevant enough.
-
Dynamic "Related Questions": After each response, the assistant suggests similar questions from a vector database, displayed in the sidebar to guide user exploration.
-
Conversation-Driven Content Generation: When a user ends a chat, a unified background process is triggered to:
- Generate FAQs: Analyzes the conversation to create and store detailed Q&A pairs in a local SQLite database.
- Generate Suggested Questions: Creates new, concise questions and adds them to Pinecone to improve future suggestions.
-
FAQ Page: A dedicated page to browse all generated FAQs, categorized for easy access.
-
Dual Database System:
- Pinecone: For the primary knowledge base and for storing suggested questions.
- SQLite: For persisting structured FAQs.
