---
title: Chat with PDF
emoji: 💬
colorFrom: indigo
colorTo: violet
sdk: gradio
sdk_version: "5.35.0"
app_file: app.py
pinned: false
---

Chat with PDF: Semantic Q&A with RAG

An interactive semantic search Q&A system that enables natural language queries over PDF documents using Retrieval-Augmented Generation (RAG).

Overview

This project implements a complete RAG pipeline for document question-answering:

• Document Processing: PDF parsing and chunking with semantic overlap
• Embedding Generation: Dense vector representations using HuggingFace models
• Vector Search: FAISS for fast similarity search across document chunks
• LLM Integration: Open-source language models (Flan-T5) for answer generation
• Conversation Memory: Multi-turn dialogue with context retention

Architecture

PDF Upload → Text Extraction → Chunking → Embedding Generation 
                                               ↓
User Query → Query Embedding → FAISS Search → Relevant Chunks
                                               ↓
                           Context + Query → LLM → Answer + Sources

Key Features

✅ End-to-end RAG pipeline built from scratch
✅ Semantic search using FAISS vector similarity
✅ Context-aware answers with source attribution
✅ Conversation memory for multi-turn interactions
✅ Production-ready deployment on HuggingFace Spaces
✅ Open-source models (no API keys required for demo)

Technical Stack

• LangChain: RAG orchestration and document processing
• HuggingFace Transformers: Embeddings and language models
• FAISS: Fast approximate nearest neighbor search
• Gradio: Interactive web interface
• Python 3.8+

Project Structure

chat-with-pdf/
├── app.py               # Entry point and Gradio app initialization
├── chain.py             # RAG pipeline: embeddings, vector store, LLM chain
├── interface.py         # UI components and event handlers
├── utils.py             # Shared state management and reset logic
├── requirements.txt     # Python dependencies
├── README.md            # Documentation (you're here!)
├── CHANGELOG.md         # Version history
└── LICENSE              # MIT license

Local Development

Installation

# Clone repository
git clone https://github.com/serasr/Chat-with-PDF.git
cd Chat-with-PDF

# Install dependencies
pip install -r requirements.txt

Run Locally

python app.py

Visit http://localhost:7860 in your browser.

Usage

1. Upload a PDF document
2. Wait for processing (embedding generation)
3. Ask questions in natural language
4. View answers with source citations
5. Ask follow-up questions (conversation context maintained)

Deploy to HuggingFace Spaces

This app is designed for easy deployment to HuggingFace Spaces:

1. Create a new Space (SDK: Gradio, SDK version: 5.35.0)
2. Upload all project files
3. Space will automatically build and deploy
4. Access at: https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME

Deployment config is managed via YAML front matter in this README.

Technical Details

RAG Pipeline Components

1. Document Processing:

• PDF text extraction using PyPDF2/pdfplumber
• Text chunking with configurable overlap (prevents context loss at boundaries)
• Metadata preservation for source attribution

2. Embedding Generation:

• Uses sentence-transformers for dense embeddings
• Supports multiple embedding models (configurable)
• Batch processing for efficiency

3. Vector Store:

• FAISS index for fast approximate nearest neighbor search
• Uses L2 (Euclidean) distance on normalized embeddings from sentence-transformers/all-MiniLM-L6-v2
• Retrieves top-2 most relevant document chunks per query (configurable via search_kwargs)

4. Answer Generation:

• Flan-T5-large for open-source generation
• Easily swappable with other HuggingFace models
• Context window management to fit relevant chunks

5. Conversation Memory:

• Tracks dialogue history
• Maintains context across multiple questions
• Configurable memory length

Performance

• Embedding Generation: ~2-5 seconds for typical PDFs (10-50 pages)
• Query Response: <2 seconds on CPU
• Memory Usage: ~2-4 GB RAM (depending on PDF size)

Roadmap

Future enhancements planned:

• Multi-PDF support (query across multiple documents)
• Embedding caching for faster re-uploads
• Highlighted source text in answers
• User feedback mechanism (thumbs up/down)
• Chat history export (Markdown/PDF)
• Multilingual support
• Advanced chunking strategies (semantic splitting)
• Model comparison (A/B testing different LLMs)

Use Cases

This RAG system is applicable to:

• Research paper analysis
• Legal document review
• Technical documentation Q&A
• Educational material comprehension
• Knowledge base search

Contributing

Contributions welcome! Areas of interest:

• Performance optimization
• Additional model integrations
• UI/UX improvements
• Documentation enhancements

Please open an issue or PR to discuss.

License

Released under MIT License. Free to use, modify, and distribute.

Acknowledgments

Built with:

• LangChain for RAG orchestration
• HuggingFace for models and deployment
• FAISS for vector search
• Gradio for UI

Contact

Sera Singha Roy

For questions about this project or my broader research on AI safety and uncertainty quantification, feel free to reach out.

---

This project demonstrates practical implementation of RAG architecture for semantic document search, a foundational technique for building retrieval-augmented LLM applications.