PrepFridge

Inspiration

Our project began with a very different idea. Initially, we wanted to create an “eBay 2.0” platform focused on reducing unnecessary electronic purchases.

For example, if someone already owned a phone but was considering buying a newer model, our platform would compare the two devices and show the price difference, meaningful feature improvements, repair options, and second-hand alternatives. The goal was to encourage users to reconsider unnecessary upgrades before creating more electronic waste.

After further discussion, we realised that the idea was too broad and mainly applied to people who frequently purchase electronic devices. We wanted to solve a problem that affects a much wider group of people.

This led us to food waste. Many people have ingredients or leftovers in their fridge but do not know what meals they can make with them. As a result, food is often forgotten until it can no longer be used. We decided to create Fridge Rescue, an application that helps users turn the food they already have into practical recipe suggestions.

What It Does

Fridge Rescue allows users to upload a photo of their fridge or remaining ingredients.

Gemini image recognition analyses the photo and identifies visible ingredients using three simple fields:

Ingredient name
Estimated quantity
Estimated urgency: fresh, mid, or close to expiry

Because image recognition may not always be accurate, we use a Human-in-the-Loop approach. Users can confirm, edit, add, or remove ingredients before continuing.

The user then answers a short preference quiz, including dietary requirements, meal type, cooking time, and the number of missing ingredients they are willing to accept.

Our backend compares the confirmed ingredients with recipes stored in our recipe database. Recipes are ranked using factors such as:

How many required ingredients the user already has
Whether the recipe uses ingredients close to expiry
How many additional ingredients are needed
Dietary compatibility
Cooking-time preferences

The application then recommends the most suitable recipes and displays recipe instructions, matched ingredients, missing ingredients, and estimated nutritional information.

How We Built It

We built the project as a web application using Next.js and TypeScript.

Gemini API is used as the vision component that converts a fridge image into structured ingredient data. We deliberately separated image detection from recipe recommendation.

Instead of asking AI to generate every recipe, we created a controlled recipe dataset containing ingredients, dietary information, cooking time, instructions, and nutrition estimates. Our backend matching algorithm searches and ranks these recipes based on the user’s confirmed ingredients and preferences.

We used AI-assisted development tools for brainstorming, prototyping, debugging, and exploring possible features. The final workflow, feature selection, database design, matching rules, and user experience were shaped through team discussions and repeated feedback.

Challenges We Faced

Our biggest challenge was controlling the project scope. We considered many features because we wanted the solution to create meaningful environmental impact. However, including too many features risked weakening the primary user journey.

We therefore focused on one core goal:

Help users cook with food they already have before it becomes waste.

Another challenge was that a single image cannot reliably identify every ingredient, estimate exact quantities, or determine whether food is safe to eat. We addressed this by making ingredient detection an initial suggestion rather than a final decision. Users remain responsible for verifying the ingredients and urgency information.

We also had to design recipe matching that prioritised waste reduction rather than simply recommending popular recipes. This led us to rank recipes more highly when they use urgent ingredients and require fewer additional purchases.

What We Learned

We learned that a meaningful product does not need the largest number of features. A focused and understandable workflow can be more valuable than a complicated platform attempting to solve several problems at once.

We also learned how AI can be combined with a structured database and deterministic backend logic. Gemini helps the application understand the image, while our own matching system determines which recipes are most suitable.

Most importantly, we learned the value of Human-in-the-Loop design. AI predictions should support the user rather than replace their judgement, especially when dealing with food freshness and safety.

What's Next

Future versions of Fridge Rescue could include user accounts, persistent fridge inventories, cooking history, saved recipes, barcode scanning, improved quantity estimation, and a dashboard showing how much food and money users have saved.

Our long-term goal is to make using existing food feel easier and more convenient than letting it go to waste.