Inspiration
Our global food system is suffering from a massive "memory leak." While 30% of all food is wasted, household budgets are being stretched to their limits. We realized that managing a kitchen is remarkably similar to managing a server, i.e. if you don't optimize your assets, the system crashes. We wanted to treat food waste not just as a social issue, but as a technical failure requiring a System Reboot
What it does
RePlate is a digital asset ledger that transforms your kitchen into a high-efficiency resource node. It monitors the "health" of your pantry, identifies "high-risk" ingredients nearing expiry, and executes a Rescue Protocol. By using AI to find the "Missing Link," it suggests recipes that maximize your GBP (£) savings while ensuring every meal passes through our Allergy Guard security layer
How we built it
We engineered RePlate as a high-performance React application using TypeScript for "Safety First" logic. The architecture functions as a modular system: The Digital Ledger: Built with React state to track inventory risk scores. Rescue Logic Engine: Powered by the Gemini 3 Flash API to generate predictive meal protocols. Safety Firewall: A filtering system that cross-references user health profiles. To quantify our efficiency, we modeled the recovery impact. If W represents total food waste and R is the recovery rate, the net impact I is summation of (W X R) - C_missing.
Challenges we ran into
The biggest "bug" we faced was a configuration error in our tsconfig.json that prevented the compiler from seeing our source code. We also hit a major roadblock with our API Authentication; initially, the system failed with INVALID_ARGUMENT errors because the API key wasn't being recognized by the Google Generative AI SDK. We had to pivot from hardcoded strings to a secure Vite Environment Variable setup using .env files and import.meta.env.
Furthermore, we encountered persistent errors with the Community Hub. Our initial "Auto-Discovery" protocol failed to reliably recognize when users were on the same network, leading to fragmented neighborhoods. We had to pivot our strategy, developing an additional manual verification section. Users now input a specific Neighborhood Access Code to initialize their local node and view nearby assets
Accomplishments that we're proud of
We are incredibly proud of our Allergy Guard implementation. Creating a system that can suggest complex proteins like Beef, Pork, and Chicken while strictly adhering to allergen constraints was a major win. We also successfully built a currency-accurate system that tracks every penny saved in GBP (£), making the environmental impact feel financially tangible.
What we learned
We learned that managing dietary preferences requires strict conditional logic rather than just creative writing. We discovered how to balance "Generative AI" with "Hard Constraints"—ensuring the AI prioritizes safety protocols over flavor alone. We also mastered the security protocols required for API key management and realized that sometimes a simple "Access Code" is more reliable for community building than complex automated network discovery.
What's next for RePlate
The next phase of our System Reboot involves:
Receipt Scanning: Simply take a photo of your grocery receipt to automatically update your kitchen inventory without manual typing.
Expanding the Network: Improving the "Neighborhood Code" system so users can share surplus food with even more people nearby.
Global Protocol: Scaling the database to support thousands of more ingredients and adding diverse currency options beyond GBP (£) to take the food rescue mission worldwide.
Log in or sign up for Devpost to join the conversation.