College Department PDF Chatbot (RAG MVP)

A Streamlit-based RAG chatbot that answers questions strictly from a fixed set of PDFs using Pinecone for retrieval and Groq for fast generation.

Features

• Hybrid Search: Combines semantic understanding with keyword matching for better retrieval
• Streaming Responses: See answers appear in real-time
• PDF-only answers: Responses are generated exclusively from your uploaded PDFs
• Source citations: Every answer includes PDF filename and page number
• Auto-indexing: PDFs are automatically indexed on first run
• Fast inference: Uses Groq for rapid LLM responses
• No hallucinations: Strict prompt engineering ensures factual answers

Tech Stack

Component	Technology
Frontend	Streamlit
Embeddings	OpenRouter (text-embedding-3-small)
Keyword Search	BM25 (pinecone-text)
Vector DB	Pinecone (hosted)
LLM	Groq API
PDF Extraction	pdfplumber

Quick Start

1. Clone and Install

git clone https://github.com/Pranavharshans/PDF-RAG.git
cd PDF-RAG
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configure Environment Variables

cp .env.example .env

Edit .env with your API keys:

• OPENROUTER_API_KEY - Get from OpenRouter
• GROQ_API_KEY - Get from Groq Console
• PINECONE_API_KEY - Get from Pinecone Console
• PINECONE_INDEX_NAME - Name for your Pinecone index

3. Create Pinecone Index

Create a serverless index in Pinecone with these exact settings:

Setting	Value
Dimensions	1536
Metric	⚠️ dotproduct (required for hybrid search)

Important: Must use dotproduct metric, not cosine!

4. Add Your PDFs

Place your PDF files in the data/pdfs/ directory.

5. Run the Application

streamlit run app.py

The application will automatically index your PDFs on first run.

Project Structure

PDF-RAG/
├── app.py                    # Main Streamlit application
├── indexer.py                # Automatic indexing logic (hybrid)
├── data/
│   ├── pdfs/                 # Place PDF files here
│   └── bm25_model.json       # Fitted BM25 model (auto-generated)
├── utils/
│   ├── __init__.py
│   ├── bm25_encoder.py       # BM25 sparse encoder
│   ├── chunking.py           # Text splitting logic
│   ├── embeddings.py         # OpenRouter embedding calls
│   ├── pinecone_utils.py     # Pinecone operations (hybrid)
│   └── pdf_loader.py         # PDF text extraction
├── requirements.txt
├── .env.example
├── GUIDE.md                  # Detailed setup guide
└── README.md

Usage

1. Open the Streamlit app in your browser
2. Type your question about the college department
3. The system will:
   • Search using hybrid search (semantic + keyword)
   • Stream the answer in real-time
   • Display source citations (PDF name + page number)

Safety Features

• Answers are generated only from indexed PDF content
• If information is not found, the system responds with a fallback message
• All answers include source citations (PDF name + page number)
• No external knowledge or speculation is used

Documentation

See GUIDE.md for detailed setup instructions and troubleshooting.

License

MIT