PDF RAG Summarizer

A Retrieval-Augmented Generation (RAG) system that allows you to summarize PDF documents and ask questions about their content. The project combines document ingestion, vector-based retrieval, and large language models to provide concise summaries and accurate, context-aware answers. It is designed as a modular, extensible Python project, suitable for coursework, research, or further development.

The system utilizes:

• LangChain as the orchestration framework,
• OpenAI for embeddings and question answering,
• Chroma as the vector database,
• Transformers for optional local PDF summarization,
• Gradio for a simple web-based user interface.

---

Installation

1. Clone the repository:

   git clone https://github.com/Flat-Earther/PDF-summary-RAG.git
   cd PDF-summary-RAG

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On macOS/Linux
   # On Windows use `venv\Scripts\activate`

3. Install the required packages:

   pip install -r requirements.txt

4. Create a .env file in the root directory and add your OpenAI API key:

   OPENAI_API_KEY=your_openai_api_key
   

5. Prepare your documents: Create a data/ directory in the project root and place your PDF files inside it.

6. Run the application:

   python app.py

7. Open your web browser and go to:

   http://localhost:7860
   

   to access the Gradio interface.

---

Project Structure

pdf_rag_project/
├── app.py                  # Application entry point (Gradio UI)
├── config.py               # Global configuration
├── ingestion/              # PDF loading, chunking, vectorstore creation
├── summarization/          # PDF summarization logic
├── rag/                    # Retrieval and question answering
├── utils/                  # Prompts and PDF export utilities
├── data/                   # Input PDF documents
├── db/                     # Chroma vector database
└── requirements.txt

---

Features

• 📄 PDF ingestion and text chunking
• 🧠 Vector-based document retrieval (RAG)
• ✍️ Automatic document summarization
• ❓ Question answering grounded in document content
• 🖥️ Simple web UI using Gradio
• 🧩 Modular architecture for easy extension

---

Important Notes

• Only PDF files are supported as input documents.

• Summarization can be done:

  • locally using Transformers (default), or
  • via an LLM fallback if configured.

• Question answering always uses retrieved document context; if the answer is not present, the model is instructed to respond with “I don’t know”.

• The default language of responses depends on the document content and prompt configuration, and can be easily modified in utils/prompts.py.