yogesh-yadav/enterprise-rag-engin

📄 Enterprise RAG Engine: Document Intelligence System

A robust, hallucination-free Retrieval-Augmented Generation (RAG) system built to perform accurate Q&A over complex or enterprise PDF documents. This application ensures high-fidelity answers by strictly grounding the AI's responses in the uploaded document context.

✨ Features

• —Intelligent PDF Ingestion: Upload any PDF to automatically extract, split, and index its contents.
• —Zero-Cost Native Embeddings: Uses the locally optimized all-MiniLM-L6-v2 Hugging Face model for textual vector embeddings.
• —Maximal Marginal Relevance (MMR) Search: Prevents context flooding by retrieving chunks that perfectly balance semantic similarity with topical diversity.
• —Deterministic AI Synthesis: Powered by Mistral AI (via LangChain) operating at temperature=0.0 to eliminate artificial hallucinations.
• —Custom Chat UI: A sleek, dark-themed chat interface with distinct user and model message layouts.

🛠️ Tech Stack

• —Frontend/UI: Streamlit
• —Framework: LangChain
• —Vector Database: ChromaDB
• —Embeddings: Hugging Face (sentence-transformers)
• —LLM: Mistral AI (mistral-small-2603)

🚀 Setup & Installation (Local & Hugging Face Spaces)

1. Prerequisites

Ensure you have Python 3.10+ installed.

2. Environment Variables

You need a Mistral API key to run the generation model.

• —Local Setup: Create a .env file in the root directory and add:

    MISTRAL_API_KEY="your_mistral_api_key_here"

• —Hingging Face Space: Go to your Space settings -> Variables and secrets -> Add a New Secret named MISTRAL_API_KEY with your actual key.

3. Dependencies

Ensure your requirements.txt includes the following (adjust versions as needed to avoid conflicts):

streamlit
python-dotenv
langchain-core
langchain-community
langchain-text-splitters
langchain-mistralai
chromadb
pypdf
sentence-transformers