DocuMind — Ask Your Documents

A RAG-powered PDF Q&A application built with LangChain, ChromaDB, and DeepSeek. Upload any PDF and ask questions grounded in its content.

🚀 Live Demo

👉 Try DocuMind

Demo

📓 Research Notebook

Full step-by-step RAG pipeline development: Open in Colab

Features

• Upload any PDF document
• Semantic search using vector embeddings
• Answers grounded strictly in the document
• Source page attribution for every answer
• Chat history within session

Tech Stack

• LangChain — RAG pipeline orchestration
• ChromaDB — Vector store for document embeddings
• HuggingFace — Embeddings (all-MiniLM-L6-v2) + DeepSeek-R1 LLM
• Streamlit — Frontend UI

Architecture

PDF → PyPDFLoader → RecursiveCharacterTextSplitter → HuggingFace Embeddings → ChromaDB
User Query → Retriever → Top-k Chunks → Prompt → DeepSeek-R1 → Answer

Setup

Clone the repo and install dependencies:

git clone https://github.com/maybedrone/DocuMind.git
cd DocuMind
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Create a .env file:

HUGGINGFACEHUB_API_TOKEN=your_token_here

Run the app:

streamlit run app.py

How It Works

1. PDF is loaded and split into chunks of 1000 characters with 200 character overlap
2. Each chunk is embedded using sentence-transformers/all-MiniLM-L6-v2
3. Embeddings are stored in ChromaDB
4. User queries are embedded and matched against stored vectors using cosine similarity
5. Top 4 most relevant chunks are retrieved and passed to DeepSeek-R1
6. Model generates an answer strictly grounded in the retrieved context