chatwithragg

PDF Chat RAG (Retrieval-Augmented Generation) TASK-1

Overview

The PDF Chat RAG is a Python-based pipeline for querying and interacting with PDF documents using Retrieval-Augmented Generation (RAG). This tool combines text extraction, chunking, vector-based retrieval, and transformer-based response generation to provide detailed and contextually accurate answers to user queries about PDF content.

---

Features

• Text Extraction: Extracts text content from PDF files using PyMuPDF (fitz).
• Text Chunking: Breaks down long texts into manageable chunks with optional overlap for contextual continuity.
• Semantic Search: Uses FAISS for efficient similarity search on text embeddings.
• Response Generation: Leverages pre-trained models like google/flan-t5-base to generate natural language responses based on retrieved text.
• End-to-End Pipeline: Enables interactive querying of PDFs with minimal setup.

---

Installation

Prerequisites

Ensure you have the following installed:

• Python 3.7+
• Pip

Required Libraries

Install dependencies using pip:

pip install numpy nltk fitz faiss-cpu sentence-transformers transformers torch

---

Usage

Running the Script

1. Place your PDF file in the working directory.

2. Update the pdf_path variable in the main() function with the path to your PDF file.

3. Run the script:

   python <script_name>.py
   

Example Queries

You can query the document with questions like:

• "Analyze the visual trends in the document."
• "Summarize the key points about data visualization."
• "Extract insights about unemployment from page 2."

---

Code Details

Pipeline Steps

1. Text Extraction: Extracts text from the PDF using the extract_text_from_pdf method.
2. Chunking: Splits text into chunks of configurable size and overlap using chunk_text.
3. Embedding and Indexing: Creates vector embeddings for the chunks using Sentence Transformers and stores them in a FAISS index.
4. Query Retrieval: Finds the most relevant chunks for the user query using semantic similarity.
5. Response Generation: Generates a natural language response using the FLAN-T5 model.

Main Classes and Methods

PDFChatRAG

• __init__(): Initializes embedding and response generation models.
• extract_text_from_pdf(pdf_path): Extracts text from the given PDF.
• chunk_text(text, chunk_size=300, overlap=50): Splits the text into manageable chunks.
• create_vector_index(chunks): Creates a FAISS index from text embeddings.
• retrieve_relevant_chunks(query, chunks, index, top_k=3): Retrieves top-K relevant chunks for a query.
• generate_response(query, retrieved_chunks): Generates a response based on the query and relevant chunks.
• chat_with_pdf(pdf_path, query): Orchestrates the entire process to handle user queries.

---

Example

def main():
    pdf_path = "data_visualization_practice.pdf"
    rag_pipeline = PDFChatRAG()

    # Example queries
    queries = [
        "Analyze the visual trends in the document",
        "Summarize the key points about data visualization",
        "Extract insights about unemployment from page 2"
    ]

    for query in queries:
        print(f"\nQuery: {query}")
        response = rag_pipeline.chat_with_pdf(pdf_path, query)
        print(f"Response: {response}")

if __name__ == "__main__":
    main()

---

Dependencies

• numpy: For numerical computations.
• nltk: For text tokenization.
• fitz: For PDF text extraction.
• faiss-cpu: For fast similarity search.
• sentence-transformers: For generating text embeddings.
• transformers: For working with transformer models.
• torch: For PyTorch-based model inference.

---

Notes

• Ensure the PDF file is text-based. Scanned PDFs may require OCR for text extraction.
• Adjust chunk_size and overlap in the chunk_text method for optimal chunking based on the document's structure.

---

Future Improvements

• Add OCR support for scanned PDFs.
• Improve response generation by fine-tuning the response model on domain-specific datasets.
• Extend support for multilingual PDFs.

---

License

This project is licensed under the MIT License.

---

Contact

For issues or suggestions, please open an issue on the project's repository.