📚 Document QA System using RAG

A Streamlit web application that lets you upload any PDF document and ask natural language questions about its contents. Powered by Retrieval-Augmented Generation (RAG) — it finds the most relevant chunks from your document and uses an LLM to generate accurate, context-grounded answers.

---

How It Works

The system follows a classic RAG pipeline:

1. Upload — You upload a PDF via the Streamlit UI.
2. Chunk — The document is split into overlapping text chunks (800 tokens, 150 overlap) using LangChain's RecursiveCharacterTextSplitter.
3. Embed — Each chunk is converted to a vector embedding using MistralAIEmbeddings.
4. Store — Embeddings are persisted in a ChromaDB vector store, keyed per uploaded file.
5. Retrieve — On each question, the top-4 most semantically similar chunks are retrieved.
6. Generate — The retrieved context is passed to ChatMistralAI (mistral-small-2506) with a strict prompt that prevents hallucination — the model only answers from the provided context.

---

Features

• Upload any PDF and build a searchable vector database with one click
• Ask free-form natural language questions
• Answers are grounded strictly in document content — no hallucinated prior knowledge
• Expandable debug panel shows exactly which chunks (with page numbers) were used to generate each answer
• Per-file vector databases — switching documents rebuilds the DB cleanly

---

Project Structure

Document-QA-System-using-RAG/
├── app.py                  # Main Streamlit application
├── main.py                 # Standalone CLI entry point
├── create_database.py      # Script to build ChromaDB from a PDF
├── document_loaders/       # Custom or extended document loader utilities
├── retrivers/              # Retriever configuration and experiments
├── Vector Store/           # Persisted ChromaDB vector store data
├── requirements.txt
└── .gitignore

---

Prerequisites

• Python 3.9+
• A Mistral AI API key

---

Installation

# 1. Clone the repository
git clone https://github.com/nandan-byte/Document-QA-System-using-RAG.git
cd Document-QA-System-using-RAG

# 2. (Recommended) Create a virtual environment
python -m venv venv
source venv/bin/activate      # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

---

Configuration

Create a .env file in the project root:

MISTRAL_API_KEY=your_mistral_api_key_here

The app loads this automatically via python-dotenv.

---

Running the App

streamlit run app.py

Then open http://localhost:8501 in your browser.

Usage:

1. Upload a PDF using the file uploader.
2. Click "Create Vector Database" and wait for processing to complete.
3. Type a question in the text input and press Enter.
4. View the AI-generated answer. Expand "🔍 Retrieved Context" to see the source chunks.

---

Tech Stack

Component	Library
UI	Streamlit
LLM	Mistral AI (mistral-small-2506)
Embeddings	langchain-mistralai MistralAI Embeddings
Vector Store	ChromaDB
PDF Loading	LangChain PyPDFLoader
Text Splitting	LangChain RecursiveCharacterTextSplitter
Orchestration	LangChain Core / Community
Env Config	python-dotenv

---

Notes

• The vector database is stored locally under chroma_db/<filename>/. Each uploaded file gets its own isolated store.
• Re-uploading the same file and clicking "Create Vector Database" will delete and rebuild the store cleanly.
• The LLM prompt enforces strict document-only answers. If your question falls outside the document's content, the model will say so rather than guess.