PDF Vector Search application

Demo video

Project Overview

This project is a PDF viewer and vector search application that can search through both images and text using HuggingFace Transformers

Features

• PDF document upload and display
• Page navigation (next, previous, jump to specific page)
• Zoom in/out functionality
• Document information display (total pages, current page)
• Indexing of text, including with OCR, and all images
• Vector search across multiple PDFs
• Text highlighting for search matches
• Sidebar for search results and navigation
• Responsive design for various screen sizes
• Persistent storage of highlights using SQLite

Technologies Used

• Next.js
• React
• TypeScript
• react-pdf library for PDF rendering
• Tailwind CSS for styling
• SQLite for local highlight storage
• HuggingFace for embedding data in preparation for vector search

Getting Started

1. Clone the repository
2. Install dependencies: pnpm install
3. Run the development server: pnpm run dev
4. Open http://localhost:3000 in your browser

Project Structure

• app/page.js: Main entry point of the application
• app/components/: React components for various parts of the application
• app/utils/: Utility functions for PDF processing and highlight storage
• app/styles/: CSS files for styling
• app/api/: API routes for handling highlight operations

Key Components

• App.tsx: Core application component
• PdfViewer.tsx: Handles PDF rendering and navigation
• KeywordSearch.tsx: Manages keyword search functionality
• HighlightPopup.tsx: Displays information about highlighted text
• Sidebar.tsx: Shows search results and navigation options
• highlightStorage.ts: Manages highlight storage operations
• clipText.ts and clipImage.ts: Handles embedding data

Features

• Has a highlight storage system supporting both SQLite
• Semantic search across images and text in multiple PDFs
• API routes for creating, retrieving, updating, and deleting highlights
• User authentication and document permissions (currently disabled)
• Export/import as JSON functionality for highlights
• Scroll the sidebar highlighted area into view across different PDFs.

Future Improvements

• Implement annotation tools (e.g., freehand drawing, text notes)
• Add support for multiple document search
• Pre-process batch PDFs for quicker highlights
• Enhance mobile responsiveness for better small-screen experience
• Optimize performance for large PDF files
• Upload the PDF into the database.

License

MIT License

Acknowledgements

• Next.js for the React framework
• react-pdf for PDF rendering capabilities
• Tailwind CSS for utility-first CSS framework
• HuggingFace for the embedding model
• Tesseract OCR for OCR output