ClaudeHopper

by Arborist-ai
Verified
# ClaudeHopper Development Roadmap This document outlines the development plan and progress for the ClaudeHopper tool, focusing on enhancing its capabilities for construction drawing search and analysis. ## Project Phases ### Phase 1: Core Text Search (Completed) -Basic infrastructure setup -LanceDB integration for vector storage -PDF text extraction and chunking -Embedding generation for text chunks -Document metadata extraction -Text search capabilities -Chunks search (specific text within documents) -Catalog search (document-level search) -Broad chunks search (search across all documents) ### Phase 2: Image Search (In Progress) -Image search API implementation -Configuration updates for image extraction -Database schema for image storage -Integration with CLIP model for text-to-image search - 🔄 Image extraction from PDFs -PDF page extraction logic - 🔄 Image embedding generation - 🔄 Metadata linking between images and source documents - 🔄 Testing with construction drawings -Performance optimization for large document sets ### Phase 3: Enhanced Visual Search (Planned) -Image-to-image similarity search -Object detection in construction drawings -Image annotation capabilities -Visual element recognition (symbols, dimensions, notes) -Advanced filtering by visual elements ### Phase 4: Integration and Advanced Features (Planned) -BIM integration capabilities -CAD file support -Multi-modal search (combining text and image queries) -Comparison tools for drawing revisions -Collaborative annotation features -UI improvements ## Current Focus We are currently focused on completing Phase 2 by implementing and testing the image search functionality. This includes: 1. Extracting images from PDF drawings 2. Generating embeddings for these images 3. Building efficient search capabilities for finding drawings based on visual similarity 4. Testing with real construction documents ## Installation Requirements for Image Search To enable image extraction from PDFs, ensure the following tools are installed: - **pdfimages** utility (part of poppler-utils) - On macOS: `brew install poppler` - On Ubuntu: `apt-get install poppler-utils` ## Next Steps 1. Test image extraction with various PDF drawing types 2. Improve the text-to-image prompt quality for better search results 3. Optimize embedding storage for large document sets 4. Add sample queries and documentation for image search ## Future Enhancements ### Phase 5: Expanded Model Support and Processing Options -Integration with hosted models through openrouter.ai or other OpenAI-compatible services - Support for GPT-4V and other advanced vision models - Configurable API settings for different model providers - Cost management and optimization features -Reprocessing capabilities - Option to rerun all analysis on existing documents - Selective reprocessing of specific documents or drawings - Batch processing improvements for large document sets - Version control for different processing runs