PDF Document Analysis and Q&A System

Problem Statement

In today's digital age, organizations and individuals often deal with numerous PDF documents containing valuable information. However, extracting specific information from multiple PDFs can be time-consuming and inefficient. Traditional methods of searching through PDFs are limited to basic keyword matching and don't understand the context or semantic meaning of the content. This creates a need for an intelligent system that can:

• Process multiple PDF documents efficiently
• Understand the context and content of the documents
• Provide accurate answers to natural language queries
• Scale with growing document collections

Solution

This project implements an intelligent PDF analysis and question-answering system that leverages Large Language Models (LLMs) and vector embeddings to provide contextual answers from PDF documents. The solution:

• Automatically processes and ingests PDF documents
• Creates semantic embeddings for efficient information retrieval
• Uses LLMs to generate human-like responses based on document content
• Provides a simple command-line interface for querying documents
• Supports multiple LLM models through Ollama integration
• Scales efficiently with document collection growth

Tech Stack

• Python: Primary programming language
• Ollama: Local LLM deployment and management
• LangChain: Framework for developing LLM applications
• Vector Database: For storing and retrieving document embeddings
• PDF Processing Libraries: For extracting text from PDF documents
• Supported Models:
  • llama2:13b
  • llama3.1:8b
  • Other Ollama-compatible models

Example

Source of Information from PDF document.

Screenshot showing the PDF with answer

Question and Answer Interface

Screenshot of the command-line interface showing a sample query and response

Setup and Installation

Step 2: Step a Virtual Environment

conda create -n venv

Step 3:Activate virtual enviroment

conda activate venv/

Step 4: Install the Requirements

pip install -r requirements.txt

Step 3: Pull the models (if you already have models loaded in Ollama, then please skip next command)

Make sure to have Ollama running on your system from https://ollama.ai

ollama pull llama3.1:8b

Step 4: put your files in the my_pdfs folder after making a directory

mkdir my_pdfs

Step 5: Ingest the files (use python3 if on mac)

python readpdf.py

Output should look like this:

Screenshot of the command-line interface showing PDF being broken into chunks and storing them in vector DBs

Step 6: Run this command (use python3 if on mac)

python Talkingpdfs.py

Play with your docs

Enter a query: How many locations does WeWork have?

Try with a different model:

ollama pull llama3.1:8b
MODEL=llama3.1:8b python Talkingpdfs.py

Screenshot of the command-line interface showing how to run the llama3.1 in local system

Add more files

Put any and all your files into the my_pdfs directory

The supported extensions are:

• .pdf: Portable Document Format (PDF),

Future Enhancements

1. Extended Document Support

• Support for additional document formats (.docx, .txt, .xlsx, .pptx)
• Handle scanned PDFs through OCR integration
• Support for documents with tables, charts, and images
• Multi-language document support and cross-language querying

2. Advanced Query Capabilities

• Multiple document comparison and cross-referencing
• Support for complex queries requiring mathematical calculations
• Document summarization features
• Topic modeling and document clustering
• Query templates for common information extraction patterns

3. Enhanced User Experience

• Web-based user interface with document visualization
• Real-time document processing status
• Query history and favorite queries
• Document annotation and highlighting
• Export functionality for answers and citations