Fridge 9000 is a mobile smart-fridge management system designed to reduce food waste by combining computer vision with inventory tracking. Users can scan their fridge with a camera, manually add products, or upload shopping receipts. A YOLO object-detection model identifies multiple products in fridge images, while receipt images and PDFs are processed using OCR. Users review detected items before changes are applied to the inventory.

The system tracks product quantities, separate expiration-date batches, partially consumed products, inventory history, and alerts for low-stock, missing, expiring, and expired food. It also includes SAM2-based product outline generation and a separate image classifier for detecting the freshness of supported foods.

Fridge 9000 also includes a human-in-the-loop ML workflow called Teach Fridge 9000. Users can correct detections or manually annotate images, with approved annotations becoming versioned training data. Candidate YOLO models can then be trained, evaluated against the active model, promoted, or rolled back while preserving the history of datasets, training runs, model versions, and evaluations.

The application is built with React Native, Expo, and TypeScript for the mobile client, and Python, FastAPI, and PostgreSQL for the backend and machine-learning pipeline, with Docker used for development and deployment.