RePlate

RePLate is your fridge's ER doctor. Snap a photo, get an AI triage of what's expiring, and cook a rescue recipe to save food before it spoils!

Comment

Inspiration

Every year, 1.3 billion tons of food are wasted globally, and the average household throws away over $1,500 worth of food annually. The root cause is what we call the "fridge blind spot"—people forget what's tucked away in the back of their fridge, don't realize what's about to expire, and struggle to come up with meals that use random ingredients before they spoil.

We were inspired by the concept of an emergency room triage system. Just as an ER doctor prioritizes patients based on urgency, RePlate prioritizes ingredients based on how soon they need to be used. The result is an AI assistant that helps rescue food before it becomes waste.

What It Does

RePlate is an AI-powered food waste reduction platform that acts like an ER doctor for your refrigerator.

Upload or Type

Users can either:

  • Camera Mode: Take a photo of their fridge or pantry.
  • Pantry Mode: Enter a list of available ingredients manually.

AI Triage

The AI analyzes the ingredients and categorizes them into three urgency levels:

  • 🔴 Rescue Tonight – Ingredients that should be used immediately.
  • 🟡 Stable – Ingredients that are still fresh for several more days.
  • ⚫ Compost – Items that appear moldy or unsafe to consume.

Rescue Recipe

The AI generates a personalized, step-by-step recipe that prioritizes the Rescue Tonight ingredients while respecting dietary preferences such as vegan or gluten-free.

Impact Tracking

Users receive:

  • A Food Waste Score (0–100)
  • An estimate of their carbon footprint savings

This helps users understand both their food waste habits and their environmental impact.

How We Built It

  • Built with Streamlit for a fast, interactive web experience.
  • Used Groq's multimodal LLM for image understanding, ingredient analysis, and recipe generation.
  • Designed structured prompts to produce consistent JSON outputs.
  • Implemented robust fallback parsing and validation to gracefully recover from malformed AI responses.
  • Customized Streamlit with extensive CSS and HTML overrides to create a polished, production-quality interface.

Challenges We Ran Into

Visual Freshness Estimation

Identifying an ingredient is relatively easy for AI, but determining whether it's fresh, overripe, or unsafe from a single image is much harder. We refined our prompts extensively and enforced conservative safety rules that always err on the side of caution.

Taming LLM Outputs

LLMs don't always return perfectly formatted JSON. We built a robust fallback parser using regex and structural matching to recover valid JSON even when responses included Markdown code blocks or extra conversational text.

UI Polish in Streamlit

Creating a startup-quality interface required extensive CSS customization and custom HTML injection to overcome Streamlit's default styling limitations.

Accomplishments We're Proud Of

A Polished User Experience

RePlate looks and feels like a real consumer product rather than a typical hackathon demo or Streamlit tutorial.

Zero-Crash Resilience

Our multi-layer fallback parser and default response system ensure the application handles API failures and malformed outputs gracefully without exposing raw errors to users.

Dual-Mode Accessibility

Supporting both Camera Mode and Pantry Mode makes the platform accessible regardless of lighting conditions, camera quality, or user preference.

What We Learned

  • How to push multimodal AI models using advanced prompt engineering and structured output schemas.
  • The importance of defensive programming when working with LLMs—never assume responses will always be perfectly formatted.
  • How to transform Streamlit's default appearance into a polished, production-ready interface through targeted CSS customization.

What's Next for RePlate

  • Receipt Scanner: Automatically add groceries to a digital pantry using OCR.
  • Smart Notifications: Remind users when ingredients are approaching expiration.
  • Community Hub: Let users discover and share successful rescue recipes.
  • Mobile App: Transition from Streamlit to React Native for a native mobile experience.

Built With

Share this project:

Updates