FridgeFlip

Turn the ingredients in your fridge into potential meals that fit your tastes and preferences. Help optimize your groceries and reduce waste.

Comment

Inspiration

According to the US Environmental Protection Agency, the average family of 4 wastes $2,913 in unused groceries each year. We wanted to reduce food waste and help create recipes utilizing what is in your fridge, as well as potential exploration with extra ingredients.

What it does

Add ingredients manually or scan your fridge with a photo. Gemini extracts photo and text uploads and into clean fields and hands it off to a kitchen council, analyzing potential meal candidates from three perspectives: taste profile, waste optimization, and nutritional value. Explore recipes tailored to your fridge and preferences. Save, skip, and rate meals, which update your user profile so the app continuously learns and adapts to your preferences.

How we built it

We built an iOS app with SwiftUI and a Python FastAPI backend. Gemini identifies ingredients from photos. Our Kitchen Council combines a learned taste model with K2-powered Nutritionist, Optimizer, and Chef roles. USDA data supports nutrition estimates, and FlavorGraph provides ingredient pairings.

Challenges we ran into

Balancing food waste reduction with meals people actually want is an ambiguous problem, so we made took into account both use-soon ingredients and preferences, while repeated skips/saves discourage/encourage different recipe ideas. Agent responses needed to be grounded in some source truth, so we pulled data from FlavorGraph and USDA nutrition in order to ensure our agent responses had actual data for scientifically proven ingredient pairings and nutritional values.

Accomplishments that we're proud of

We built a working app that turns fridge ingredients into recipe suggestions through our Kitchen Council, where recommendations are continuously adapting and improving to the user.

What we learned

We learned about using LLM APIs, agent orchestration, and agent architecture during the project. We incorporated user feedback loops to constantly update the user profile which gives our kitchen council agents new inputs to continuously adapt to their tastes.

What's next for FridgeFlip

We want to test with real users, expand ingredient and nutrition coverage, and connect to actual grocery stores around the user for accurate price and additional ingredient purchasing options.

Built With

Share this project:

Updates

Submission history