Cosmic Cooler

Upload a photo of your fridge and an AI will tell you what zodiac sign it is. Built as a tongue-in-cheek astrology app — "astrology for people who work in spreadsheets."

How It Works

1. Upload a photo of your fridge (or skip to get a random reading)
2. ML5.js classifies the image using a MobileNet-based model trained on fridge photos
3. Get your result — a zodiac sign, confidence score, personality blurb, and the chance to tell us if we got it right

If no trained model is available, it falls back to random cosmic intuition.

Tech Stack

• SvelteKit (Svelte 5 with runes)
• TailwindCSS 4
• ML5.js 0.5.0 — transfer learning with MobileNet FeatureExtractor
• Vite

Getting Started

npm install
npm run dev

Open http://localhost:5173.

Scripts

Command	Description
npm run dev	Start dev server
npm run build	Production build
npm run preview	Preview production build
npm run lint	Run ESLint + Prettier checks
npm run format	Auto-format with Prettier

Training Your Own Model

1. Add fridge images to static/images/{zodiac-sign}/ folders
2. Update static/dataset.json with image paths and labels
3. Visit /train to run transfer learning in the browser
4. Export the model and place files in static/models/fridge-zodiac/

Project Structure

src/
├── lib/
│   ├── assets/        # Images, icons, zodiac sign SVGs
│   ├── components/    # Shared UI components (Drop, Pill, etc.)
│   ├── zodiac.js      # Zodiac sign data and utilities
│   └── index.js       # Sign definitions and personality copy
├── routes/
│   ├── +page.svelte        # Home — upload your fridge photo
│   ├── loader/+page.svelte # Loads ML5 model and classifies image
│   ├── lezgo/+page.svelte  # Results page with prediction
│   └── train/+page.svelte  # Model training interface
└── app.html

static/
├── images/    # Training images by zodiac sign
├── models/    # Exported ML5 model files
└── dataset.json