FRIDGe
Inventory-aware cooking with validated AI suggestions
A mobile kitchen-management prototype that tracks ingredient quantities and freshness, supports photo-based item entry, and validates recipe suggestions against the household’s actual inventory.
PROJECT OVERVIEW
FRIDGe connects a household’s kitchen inventory to the question of what to cook next. The app includes ingredient entry and freshness tracking, a photo-scan review flow, recipe suggestions and details, and guided Cook Mode. Quantities are represented numerically in grams, milliliters, or pieces, with a catalog used for ingredient matching and supported conversions.
THE PROBLEM
A useful recipe recommendation must respect both the ingredients available and their quantities. Plausible AI suggestions can still invent an ingredient or request more than a household has. FRIDGe treats generated recipes as proposals that need validation before they can be shown or stored.
TECHNICAL APPROACH
An Expo client uses Zustand stores and an authenticated Express API backed by Prisma and PostgreSQL. OpenAI requests stay on the server and return structured JSON. A validation layer compares recipe ingredients and quantities with real inventory, retries once with concrete violations, and drops recipes that remain invalid. Calorie estimates are computed as ranges from catalog data, and cooking deductions use transactional inventory updates.
PROJECT OUTCOME
Implemented the inventory, recipe-validation, calorie-estimation, scan-review, and cooking workflows in a local prototype. Live verification of OpenAI-dependent scan and suggestion flows remains blocked by API billing, and a complete on-device walkthrough and production deployment are still pending.