RAGForge

Overview

RAGForge is a fully local, open-source Retrieval-Augmented Generation (RAG) application. It lets you upload multiple documents — PDFs, text files, or Markdown — and have a multi-turn AI conversation about their contents.

Unlike generic chatbots, RAGForge strictly grounds every answer in your documents, shows you the exact source pages it used, tracks token costs, and prevents hallucinations.

Key Features

• ▸Hybrid Search — BM25 keyword + ChromaDB vector similarity via EnsembleRetriever
• ▸Query Expansion — LLM generates sub-queries to improve document recall
• ▸Cost & Token Tracker — Live token count + USD cost estimate per response
• ▸Hallucination Guard — Post-generation validator rejects out-of-context answers
• ▸Docker Ready — One-command container deployment with health checks

Technical Stack

• ▸Backend/AI: Python, LangChain, ChromaDB
• ▸UI: Streamlit
• ▸AI Models: Google Gemini 1.5/2.5 Flash, HuggingFace embeddings