🧠 DocuMind AI

AI-Powered Academic Knowledge Assistant
Retrieval-Augmented Generation (RAG) System
MCA Prompt Engineering Assignment — Chanakya University

---

📌 What is DocuMind AI?

DocuMind AI is a production-quality RAG (Retrieval-Augmented Generation) system that lets you upload any PDF or TXT document and ask questions about it. The AI answers strictly from your uploaded documents — no hallucinations, no guessing.

Built as a Prompt Engineering assignment demonstrating the full RAG pipeline:

User Query → Embed Query → Vector Search → Retrieve Chunks → LLaMA 3 → Grounded Answer

---

✨ Features

• 📄 Upload PDF and TXT documents
• 🔍 Semantic similarity search using all-MiniLM-L6-v2
• 🧠 Grounded answers using LLaMA 3 8B Instruct via OpenRouter
• 💾 Persistent ChromaDB vector store
• 📚 Source attribution — see exactly which chunks were used
• 💬 Multi-turn chat with history
• 🎨 Premium Streamlit UI with dark mode toggle
• ⚡ Cached models for fast loading

---

🏗️ Architecture

rag_project/
│
├── app.py                  ← Main Streamlit application
├── requirements.txt        ← Python dependencies
├── .env.example            ← Environment variable template
├── .gitignore
├── README.md
│
├── rag/
│   ├── pdf_loader.py       ← Document loading & text extraction
│   ├── chunker.py          ← Overlapping text chunking
│   ├── embeddings.py       ← Sentence-transformer embeddings
│   ├── vector_store.py     ← ChromaDB vector database
│   ├── retriever.py        ← Semantic similarity retrieval
│   ├── llm.py              ← OpenRouter API caller
│   └── rag_pipeline.py     ← Orchestrates full RAG pipeline
│
└── assets/
    └── styles.css          ← Premium UI stylesheet

RAG Pipeline (5 Stages)

Stage	Module	Description
1. Ingestion	pdf_loader.py	Extract clean text from PDF/TXT
2. Chunking	chunker.py	Split into 500-char overlapping chunks
3. Embedding	embeddings.py	Convert text to 384-dim vectors
4. Storage	vector_store.py	Persist in ChromaDB with cosine similarity
5. Generation	rag_pipeline.py	Retrieve + augment + generate answer

---

🚀 Quick Start

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/documind-ai.git
cd documind-ai

2. Create a virtual environment

python -m venv venv

# Windows
venv\Scripts\activate

# macOS/Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

⏳ First install downloads the embedding model (~90 MB). This only happens once.

4. Set up your API key

# Copy the example file
cp .env.example .env

Edit .env and add your key:

OPENROUTER_API_KEY=your_key_here

Get a free API key at openrouter.ai/keys — no credit card required for free tier.

5. Run the app

streamlit run app.py

Open http://localhost:8501 in your browser.

---

🖥️ Usage

1. Upload documents using the sidebar file uploader (PDF or TXT)
2. Wait for indexing to complete (shown in sidebar)
3. Type a question in the chat bar at the bottom
4. View the grounded answer and expand Retrieved Sources to see evidence

---

🔧 Configuration

Parameter	Default	Location
Chunk size	500 chars	rag/chunker.py
Chunk overlap	50 chars	rag/chunker.py
Top-K retrieval	5 chunks	rag/rag_pipeline.py
LLM model	meta-llama/llama-3-8b-instruct	rag/llm.py
Embedding model	all-MiniLM-L6-v2	rag/embeddings.py
Temperature	0.2	rag/llm.py

---

📸 Screenshots

🤖 Tech Stack

Component	Technology
Frontend + Backend	Streamlit
Vector Database	ChromaDB
Embedding Model	all-MiniLM-L6-v2
LLM API	OpenRouter (LLaMA 3 8B Instruct)
PDF Parsing	PyPDF
Environment	python-dotenv

---

📚 References

• Lewis et al. (2020) — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
• Course: Prompt Engineering — Chanakya University, School of Engineering
• Instructor: Mr. Deepak B
• Student: Baire Gowda

---

📄 License

MIT License — free to use for academic and personal projects.