An AI-powered web application that answers questions about your PDF documents using Retrieval-Augmented Generation (RAG), just like ChatGPT for PDFs!
Ask questions about any PDF document and get instant AI answers! This app gives you:
- Smart Document Analysis 📄 - Upload any PDF and get instant insights
- Conversational AI 💬 - Ask follow-up questions with memory context
- Source Citations 🔍 - See exact page numbers where answers come from
- Demo Mode 🚀 - Try it instantly with pre-loaded college guide PDF
Perfect for students, researchers, and professionals who work with documents.
- Instant PDF Processing - Upload and analyze documents in seconds
- AI-Powered Answers - Uses Meta-Llama-3-8B for intelligent responses
- Memory System - Remembers conversation context for follow-ups
- Source Tracking - Shows page numbers and relevance scores
- Professional UI - Clean interface with progress indicators
- Chat History - Download conversation records as text files
Document Processing:
- Splits PDF into smart text chunks (1000 chars with 200 overlap)
- Creates vector embeddings using sentence-transformers
- Stores in FAISS database for fast similarity search
Question Answering:
- Finds most relevant document sections using vector similarity
- Sends context to Meta-Llama-3-8B via HuggingFace API
- Maintains conversation memory for contextual responses
- Provides page citations with confidence scores
- Python - Core programming language
- Streamlit - Web application framework
- LangChain - RAG pipeline orchestration
- FAISS - Vector similarity search
- HuggingFace - AI model API integration
- Sentence Transformers - Text embeddings
- PyPDF - PDF text extraction
- Meta-Llama-3-8B - Large language model
SmartPDF-AI/
├── rag_app.py # Main Streamlit application
├── rag_system.ipynb # Development notebook
├── demo_pdf.pdf # Sample PDF for testing
├── requirements.txt # Dependencies
├── app_screenshot.jpg # UI screenshot
└── README.md # Documentation
Document Processing:
- Text chunking: 1000 characters with 200 overlap for context
- Embeddings: all-MiniLM-L6-v2 model (90MB, fast CPU inference)
- Vector search: FAISS with cosine similarity ranking
AI Response:
- Context retrieval: Top 5 relevant chunks per query
- Response time: 3-8 seconds including API latency
- Memory system: Tracks last 4 conversation turns
- Accuracy: High relevance with source page citations
- Clone repository
git clone https://github.com/sankaran-s2001/SmartPDF-AI-RAG-Based-Document-Q-A-System.git
cd smartpdf-ai-rag- Install dependencies
pip install -r requirements.txt- Get HuggingFace API token
- Visit HuggingFace Settings
- Create token with "Read" permission
- Copy token (starts with
hf_)
- Run application
streamlit run rag_app.pyAI/ML Engineering:
- RAG (Retrieval-Augmented Generation) architecture
- Vector databases and similarity search algorithms
- LLM integration with API-based inference
Full-Stack Development:
- Streamlit web application with custom CSS styling
- Session state management and memory systems
- File upload and processing workflows
Production Deployment:
- Dependency optimization for cloud deployment
- Error handling and user experience design
- Real-time progress tracking and status updates
- ✅ Built production-ready RAG system from scratch
- ✅ Integrated multiple AI technologies (embeddings, vector DB, LLM)
- ✅ Implemented conversation memory with context awareness
- ✅ Created professional UI with progress indicators
- ✅ Optimized for cloud deployment (400MB total dependencies)
- ✅ Added comprehensive error handling and fallbacks
- Multi-PDF support for document collections
- Advanced chunking strategies (semantic splitting)
- User authentication and personal document storage
- Export answers as formatted reports
- Mobile-responsive design
Sankaran S
Demonstrating AI engineering skills through intelligent document processing. Perfect for anyone working with PDF documents!
⭐ If you find this helpful, please star the repository!