Daft Image Search Tool - Product Requirements Document

Project Overview

We’re creating an intelligent image search tool using Daft.ai for efficient data processing, Flask for the backend API, and jQuery for the frontend interface.

Technical Stack

• Backend: Flask (Python)
• Frontend: jQuery, Bootstrap, HTML5
• Data Pipeline: Daft.ai for distributed processing
• AI Models: BLIP (Salesforce) for image captioning and tagging
• Image Processing: Pillow for resizing and format handling

Core Features

Data Loader Page

Purpose: Process and catalog images from local folders

Features:

• Allow users to specify local folder paths for image processing
• Recursively discover all images in folders and subfolders
• Use Daft.ai to efficiently batch process large image collections
• Resize all images to standardized dimensions (224x224) for consistency
• Run each image through BLIP AI model for automatic captioning
• Extract relevant tags from generated captions and filenames
• Generate structured JSON database with image metadata
• Provide real-time progress tracking during processing
• Handle multiple image formats (JPEG, PNG, GIF, WEBP, BMP, TIFF)

Technical Requirements:

• Parallel processing using Daft.ai DataFrames
• Background job management with status tracking
• Error handling for corrupted or unsupported files
• Memory-efficient processing for large datasets
• Comprehensive metadata extraction (file size, dates, dimensions)

Image Library Page

Purpose: Search and browse processed image collections

Features:

• Text-based search using natural language queries
• Tag-based filtering and discovery
• Responsive grid layout for image browsing
• Modal view for detailed image inspection
• Search result statistics and sorting
• Mobile-friendly responsive design
• Fast search performance with client-side optimization

Technical Requirements:

• RESTful API for search operations
• Efficient JSON-based data querying
• Lazy loading for performance optimization
• Semantic search matching tags and captions
• Progressive image loading with thumbnails

Data Schema

{
  "metadata": {
    "processed_date": "ISO timestamp",
    "source_folder": "original folder path",
    "total_images": "number of processed images"
  },
  "images": [
    {
      "id": "unique identifier",
      "filename": "original filename",
      "original_path": "full path to original",
      "processed_path": "path to resized image",
      "file_size": "size in bytes",
      "created_date": "file creation date",
      "tags": ["array", "of", "extracted", "tags"],
      "caption": "AI-generated description",
      "processed_date": "processing timestamp"
    }
  ]
}

API Endpoints

• POST /api/process - Start image processing job
• GET /api/jobs/{id} - Check processing status
• POST /api/search - Search images by text query
• GET /api/images - Retrieve all processed images
• GET /processed_images/{filename} - Serve processed image files

User Experience Goals

• Simplicity: Intuitive interface requiring minimal user training
• Performance: Fast processing and search response times
• Reliability: Robust error handling and progress feedback
• Scalability: Handle large image collections efficiently
• Accessibility: Responsive design working across devices

Success Metrics

• Process 1000+ images in under 5 minutes
• Search response time under 2 seconds
• Support for all common image formats
• 90%+ accuracy in automatic tagging
• Mobile-responsive interface