Skip to content

Latest commit

 

History

History
100 lines (62 loc) · 2.76 KB

README.md

File metadata and controls

100 lines (62 loc) · 2.76 KB

DocChat: Chat With PDF

DocChat is a Streamlit-based chatbot application that allows you to interact with PDF documents in a conversational manner. It leverages a locally stored sentence-transformer model for semantic search and a local GPT model for generating contextual responses.

Features

  • Upload PDFs: Easily upload PDF documents.
  • Semantic Search: Search the document content using state-of-the-art embeddings.
  • Conversational Responses: Ask questions and get meaningful responses from the PDF's content.
  • Completely Offline: No API usage; models are stored and run locally.

DocChat Preview

DocChat Preview

Installation and Setup

1. Install Required Dependencies

Ensure you have Python 3.8 or later installed. Then, install the required dependencies:

pip install -r requirements.txt

2. Download and Install Models Locally

Before running the application, download the necessary models locally by running the model-install.py script:

python model-install.py

This script downloads and stores:

  • The SentenceTransformer model for semantic search.
  • The GPT-Neo model for conversational response generation.

4. Start the Application

Once the models are downloaded, run the Streamlit app to start DocChat:

streamlit run app.py

Usage

Step 1: Upload a PDF

  • On the main interface, click on the Upload PDF button to upload your document.

Step 2: Ask Questions

  • Use the chat interface to ask questions about the uploaded PDF.
  • Your messages and the bot's responses will appear in a conversational format.

Step 3: View Chat History

  • The app maintains a history of your conversation with the document for reference.

Project Structure

.
├── app.py               # Main Streamlit application
├── model-install.py     # Script to download and set up models locally
├── requirements.txt     # List of required Python packages
├── README.md            # Project documentation
├── models/              # Directory for storing downloaded models

How It Works

  1. Semantic Search:

    • The PDF is parsed into chunks of text using the SentenceTransformer model for embeddings.
    • FAISS (a similarity search library) is used to perform efficient nearest-neighbor searches.

  2. Contextual Response Generation:

    • Retrieved chunks are passed as context to the GPT model.
    • GPT generates a concise response tailored to your query.

Notes

  • Ensure sufficient disk space for storing models (~1GB).
  • For better performance, larger GPT models like GPT-2 or GPT-J can be used.
  • If you encounter errors, ensure all dependencies are correctly installed and the models are downloaded without interruptions.