Chat with PDFs

This repository contains a streamlit application that allows a user to chat with their PDF documents. The user can ask a question about the document's contents in natural language, and the application will provide a response based on the content. A language model is used to generate answers.

Contents

• Architecture
• Requirements
• Usage
• Note

Architecture

The application follows these steps to provide responses to your questions:

• Text Extraction: Extracts text from the uploaded PDF documents.
• Text Chunking: The extracted text is divided into manageable chunks.
• LLM: The app uses a language model to generate vector representations - embeddings of the text chunks.
• Semantic Matching: When the user asks a question, the app compares it with the text chunks and identifies the semantically similar chunks.
• Response: The similar chunks are passed to the language model, which generates a response based on the content of the PDFs.

The model used in this project is google's flan-t5-xxl

Requirements

• langchain
• PyPDF2
• python-dotenv
• streamlit
• faiss-cpu
• altair
• tiktoken
• huggingface-hub
• InstructorEmbedding
• sentence-transformers
• torch
• torchvision
• setuptools

pip install langchain PyPDF2 python-dotenv streamlit faiss-cpu altair tiktoken huggingface-hub InstructorEmbedding sentence-transformers torch torchvision setuptools

Obtain an API Token from HuggingFace and add it to the .env file in your directory

HUGGINGFACEHUB_API_TOKEN = your_api_key

Usage

• Install the required dependencies.
• Run app.py to start the streamlit application, using

streamlit run app.py

• Upload your PDF documents using the file uploader.
• Click the Process button to process the uploaded documents.
• Ask questions through the chat.

Note

This project uses Instructor to generate text embeddings and runs locally on your machine. The processing time will vary based on your hardware.