Category Image App

A .NET web application that integrates with RavenDB and OpenAI's CLIP model for category management with image embeddings.

Features

• Category Management: Create, view, and search categories
• Image Support: Upload and display images for categories stored as RavenDB attachments
• Text Search: Traditional text-based search through category names and descriptions
• Semantic Search: AI-powered search using OpenAI embeddings
• Embedding Generation: Generate embeddings for categories using OpenAI's CLIP model
• Modern UI: Bootstrap-based responsive interface

Prerequisites

1. RavenDB: Install and run RavenDB locally on port 8080
2. OpenAI API Key: Get an API key from OpenAI
3. .NET 8 SDK: Install .NET 8 SDK

Setup Instructions

1. RavenDB Setup

• Download RavenDB from https://ravendb.net/download
• Run RavenDB locally (default port 8080)
• The application will automatically create the database "CategoryImageDB"

2. OpenAI Configuration

• Get your OpenAI API key from https://platform.openai.com/api-keys
• Update the API key in appsettings.json or appsettings.Development.json:

{
  "OpenAI": {
    "ApiKey": "your-actual-openai-api-key-here"
  }
}

3. Running the Application

# Restore packages
dotnet restore

# Run the application
dotnet run

The application will be available at https://localhost:5001 or http://localhost:5000.

Usage

Initial Setup

1. Create Sample Data: Click "Create Sample Data" to generate some example categories
2. Initialize System: Click "Initialize System" to generate embeddings for all categories using OpenAI's CLIP model

Features

• Browse Categories: View all categories with their images
• Text Search: Use the search bar for traditional keyword search
• Semantic Search: Check "Use AI Semantic Search" for AI-powered search
• Add Categories: Click "Add Category" to create new categories with optional images

Architecture

Models

• Category: Main entity with name, description, image support, and embeddings
• SearchRequest: DTO for search operations

Services

• CategoryService: Handles CRUD operations and similarity search
• OpenAIService: Integrates with OpenAI API for embeddings
• EmbeddingService: Orchestrates embedding generation

Database

• RavenDB: Document database with attachment support
• Indexes: Custom indexes for text search and embedding operations

API Endpoints

• GET / - Main categories page with search
• GET /Home/CreateCategory - Category creation form
• POST /Home/CreateCategory - Create new category
• POST /Home/InitializeSystem - Generate embeddings for all categories
• POST /Home/CreateSampleData - Create sample categories
• GET /Home/GetCategoryImage/{id} - Retrieve category image

Technical Details

Embedding Process

1. For categories with images: Generate embeddings from the image using OpenAI's CLIP model
2. For categories without images: Generate embeddings from text (name + description)
3. Store embeddings in RavenDB for similarity search

Search Types

1. Text Search: Uses RavenDB's full-text search capabilities
2. Semantic Search: Uses cosine similarity with OpenAI embeddings

Image Storage

• Images are stored as RavenDB attachments
• Supports various image formats (JPEG, PNG, etc.)
• Images are displayed directly from RavenDB

Configuration

Key configuration options in appsettings.json:

{
  "OpenAI": {
    "ApiKey": "your-openai-api-key"
  },
  "RavenDB": {
    "Urls": ["http://localhost:8080"],
    "Database": "CategoryImageDB"
  }
}

Dependencies

• RavenDB.Client: Document database client
• Microsoft.AspNetCore.Mvc.Razor.RuntimeCompilation: For development-time view compilation
• Newtonsoft.Json: JSON serialization
• Bootstrap 5: UI framework
• Font Awesome: Icons

Notes

• The application uses OpenAI's text-embedding-3-large model
• Rate limiting is implemented to avoid OpenAI API limits
• Embeddings are cached in RavenDB to avoid regeneration
• The UI is responsive and works on mobile devices