Fridge Raid

Website | Proposal | Technical Documentation | Developer

Inspiration

People often have ingredients sitting in their fridge but don't know what to make with them. They also usually have cookbooks gathering dust on the shelf, but it takes a long time to cross reference this with what ingredients you already have.

We wanted to build an app that flips the traditional recipe search on its head: instead of finding a recipe and then buying ingredients, you start with what you already have and discover what you can cook.

What it does

Fridge Raid lets users add ingredients they have at home and instantly matches them to 500 recipes (and counting). It suggests recipes ranked by how many matching ingredients you already have, lets you save favorites to a personal Cookbook, and generates a Shopping List for any missing ingredients. Users can also import recipes by scanning a photo (OCR) or pasting a URL, and switch between original, metric, and imperial unit systems.

How we built it

We built it with React Native and Expo, using a tab-based navigation structure. The recipe database was sourced from various free APIs and refined extensively by hand — normalizing ingredients, adding alternative names, assigning categories, difficulty levels, and meal types. For recipe scanning and URL imports, we integrated Amazon Bedrock's Nova Pro model to parse and format recipes using LLM-powered OCR. We used RevenueCat for subscription management and AsyncStorage for local data persistence.

Challenges we ran into

Ingredient matching was tricky — the same ingredient can be called many different things (e.g., "coriander" vs "cilantro"), so we had to build a matching system with alternative names and case-insensitive lookups.
Parsing fractional ingredient quantities (like "1 1/2 cups") required careful handling.
Recipes with "or" ingredients (e.g., "butter or margarine") needed special logic to let users choose.
Normalizing 500 recipes from raw API data into a consistent format with proper units, preparation notes, and categories was a massive data-cleaning effort across many iterations.

Accomplishments that we're proud of

A database of 500 hand-refined recipes spanning 20 cuisines and 14 meal types.
LLM-powered recipe import — users can snap a photo of a recipe card or paste a URL and it gets automatically parsed into the app's format.
A unit conversion system that can switch entire recipes between original, metric, and imperial measurements.
Smart recipe suggestions that rank results by ingredient match percentage so users always see the most cookable recipes first.

What we learned

Data quality matters more than data quantity — we spent more time refining ingredient names, alternative names, and recipe metadata than writing app code.
LLMs are powerful for unstructured data parsing (OCR, URLs) but need carefully crafted prompts to return consistent, structured output.
Building a good ingredient matching system requires thinking about how real people describe food, not just exact string matches.

What's next for Fridge Raid

Expanding the recipe database with more cuisines and dietary options (vegetarian, vegan, gluten-free filters).
Adding meal planning features so users can plan their week and generate a consolidated shopping list.
Community recipe sharing so users can publish their own imported recipes.
Smarter suggestions using past cooking history and user preferences.
User uploaded images stored for recipes.
Recipe step images showing how to do the cooking, matched to common steps even in custom recipes.
Selectable color scheme
Rotating sponsored chefs, suggesting their own recipes the user would like.