ChatDoc-AI (🧠 Conversational RAG Q&A System with PDF Uploads and Chat History)

This project implements a Retrieval-Augmented Generation (RAG) pipeline that allows users to upload PDF files and chat with their content through a conversational interface.
It maintains chat history, reformulates follow-up questions contextually, and retrieves accurate answers using semantic search and LLM reasoning.

---

🚀 Features

• 📄 Upload Multiple PDFs: Ingest and process multiple documents at once.
• 🧩 Automatic Chunking: Splits long text into overlapping chunks for better context handling.
• 🔍 Context-Aware Retrieval: Retrieves the most relevant document chunks using semantic similarity.
• 💬 Conversational Memory: Maintains session-based chat history for coherent multi-turn conversations.
• ⚙️ RAG Pipeline: Combines retriever + LLM for accurate and context-grounded answers.
• 🔑 Groq LLM Integration: Uses llama-3.1-8b-instant model for fast, high-quality responses.
• 🧠 Standalone Question Rewriting: Reformulates follow-up questions into self-contained ones for improved retrieval.
• 🎨 Streamlit UI: Simple and interactive interface for chatting with PDFs.

---

🧰 Tech Stack

Component	Technology
Framework	Streamlit
LLM	Groq API – llama-3.1-8b-instant
Vector Store	Chroma
Embeddings	Hugging Face – all-MiniLM-L6-v2
Document Loader	PyPDFLoader (LangChain)
Memory	ChatMessageHistory (LangChain)
Environment	Python 3.10+, dotenv

---

🧱 Project Structure

📦 Conversational-RAG
│
├── app.py # Main Streamlit app
├── requirements.txt # Dependencies
├── .env # Environment variables (HF_TOKEN)
└── README.md # Project documentation

---

---

⚙️ Installation & Setup

🪄 Option 1: Fork & Run This Project

If you want to run your own copy of this project:

1. Fork the Repository

   • Go to the GitHub repository:
     👉 https://github.com/avanigupta06/ChatDoc-AI
   • Click the “Fork” button (top-right corner) to create your own copy.

2. Clone Your Fork

   git clone https://github.com/<your-github-username>/ChatDoc-AI.git
   cd ChatDoc-AI

3. Create a Virtual Environment

   python -m venv venv
   venv\Scripts\activate       # On Windows
   source venv/bin/activate    # On macOS/Linux

4. Install Dependencies

   pip install -r requirements.txt

5. Set Up Environment Variables

   Create a .env file in the project root and add:
   
   HF_TOKEN=your_huggingface_api_token

6. Run the App

   streamlit run app.py

Open in Browser Visit the local URL shown in the terminal (usually http://localhost:8501 ).

---

🔑 API Keys Required

You’ll need two keys to run this project:

API	Purpose	Where to Get It
Hugging Face Token	Embeddings (all-MiniLM-L6-v2)	https://huggingface.co/settings/tokens
Groq API Key	Access to Llama 3.1 model	https://console.groq.com/keys

You can input the Groq API Key directly in the Streamlit interface when prompted.

---

💡 How It Works

• Upload PDFs → PDFs are parsed into text using PyPDFLoader.
• Chunk Text → Long text is split into 5000-character chunks with 500 overlap.
• Generate Embeddings → Each chunk is converted into a numerical vector using Hugging Face embeddings.
• Store in Chroma DB → The embeddings are stored for semantic retrieval.
• Ask a Question → The LLM reformulates the question if needed and retrieves relevant chunks.
• Answer Generation → Llama-3.1 uses retrieved context to generate concise, grounded answers.
• Chat Memory → Conversation history is maintained per session for contextual continuity.

---

📸 Screenshots

---

If you find this project useful, please ⭐ star this repository on GitHub!

---