RAG-BASED-CHATBOT

The RAG Knowledge Assistant is a modular Retrieval-Augmented Generation (RAG) system where each component has a single responsibility. The system separates ingestion, retrieval, routing, generation, and the API/UI layer, making them fully decoupled and independently testable. It uses a query router that classifies incoming queries as either general or document-based before retrieval. For general queries, a plain LLM call is used, while document queries trigger retrieval from a Chroma vector store. The retrieved context is formatted with sources and passed to a RAG prompt and LLM. The final output is parsed into structured JSON containing an answer and citations. The tech stack includes LangChain (LCEL) for orchestration, GPT-5.6 Luna via an OpenAI-compatible gateway for generation, OpenAI text-embedding-3-small for embeddings, ChromaDB as the vector store, Pydantic for structured output, FastAPI for the backend, and Streamlit for the frontend. Evaluation is done with RAGAS, measuring faithfulness, answer relevancy, and context precision.