Document QA System

This Streamlit application is designed to provide a Question-Answering (QA) system for PDF documents. It leverages Optical Character Recognition (OCR) to extract text from uploaded PDF files and uses a pre-trained model from Hugging Face's transformers library to answer questions based on the extracted text.

Features

• PDF Upload: Users can upload PDF documents to the application.
• PDF to Image Conversion: The application converts PDF pages into images.
• OCR Processing: Extracts text from the converted images using pytesseract.
• Question Answering: Users can ask questions based on the extracted text, and the app uses a pre-trained QA model to provide answers.

Requirements

To run this application, you need the following libraries:

• streamlit: For creating the web application.
• transformers: From Hugging Face, used for the QA model.
• pdf2image: For converting PDF pages to images.
• pytesseract: For OCR capabilities.
• Pillow: For image processing.
• tempfile: For handling temporary files.
• os: For operating system dependent functionalities.

Installation

1. Clone the repository:

   git clone https://github.com/manavkdubey/document-qa

2. Install dependencies: You can install the necessary libraries using pip. It's recommended to do this in a virtual environment:

   pip install streamlit transformers pdf2image pytesseract Pillow

3. Install Tesseract-OCR: Pytesseract requires the Tesseract-OCR engine. Follow the installation instructions specific to your operating system.

Usage

1. Run the Streamlit app: In the project directory, execute:

   streamlit run app.py

2. Upload PDF Document: Use the file uploader to upload a PDF document.

3. Extract Text: The app automatically converts the PDF to images, performs OCR, and extracts the text.

4. Ask Questions: Enter your question in the provided text input field. The app will display the answer based on the context of the extracted text.

Note

• The performance of the OCR depends on the quality of the PDF images.
• The accuracy of answers depends on the pre-trained QA model from Hugging Face.

Contributions

Contributions are welcome. Please fork the repository and submit pull requests with your enhancements.

License

This repository is under MIT License