Rescue Chef AI — Zero Waste. Maximum Flavor.

Snap your fridge. Get a recipe. Missing ingredients? AI scores safe substitutions or finds the nearest store. Cook smarter, waste less. Five Gemini APIs, zero food waste. One tap.

Comment

Inspiration

Every year, households throw away a third of the food they buy — not because it's spoiled, but because they don't know what to cook with what's left. Rescue Chef AI was born from a simple question: what if your phone could look inside your fridge and turn forgotten ingredients into a real meal?

What it does

Rescue Chef AI is a kitchen intelligence app that fights food waste through two AI-powered paths: 📸 Scan → Decide → Cook

  1. Scan your fridge — Gemini 3 Flash with Vision identifies every ingredient from a photo
  2. Get a rescue plan — The AI generates a full recipe and compares it against what you have
  3. Choose your path:
    • 🔧 HACK IT — AI substitutions scored for taste match, swap safety, and texture (e.g., "Use yogurt instead of cream — 92% safe")
    • 🛒 SHOP IT — A precise shopping list with estimated cost and nearby stores via Google Maps

Each recipe includes an AI-generated watercolor illustration you can edit with text prompts, step-by-step cooking with a timer, and text-to-speech for hands-free guidance.

How we built it

Built entirely on Google AI Studio using Vibe Coding (no-code). The app chains five Gemini capabilities in a single flow:

Capability Model Purpose
Vision Gemini 3 Flash Ingredient detection from fridge photos
Thinking Mode Gemini 3 Flash Recipe generation + substitution scoring via structured JSON
Maps Grounding Gemini 2.5 Flash Nearby grocery stores with real-time status
Nano Banana Gemini 2.5 Flash Recipe illustration generation + live text-based editing
TTS Gemini 2.5 Flash Hands-free cooking step narration

The decision engine uses structured JSON output with a detailed schema covering ingredients, hacks, safety factors, shopping lists, and environmental impact — all in a single API call with Thinking Mode enabled.

Challenges we ran into

  • Emoji rendering — Gemini sometimes returned emoji names ("thermometer") instead of Unicode characters. Solved with a dual approach: prompt constraints + client-side fallback mapping
  • localStorage overflow — Base64 recipe images exceeded the 5MB browser limit. Fixed by stripping image data before persistence and regenerating on demand
  • Safety score confusion — Early versions used technical food science jargon ("Cross-Contamination", "Acidity Balance") that confused testers. Replaced with user-friendly categories: Taste Match, Swap Safety, Texture Match
  • Maps API constraints — Google Maps grounding only works on Gemini 2.5 Flash, not Gemini 3. Required a hybrid model architecture

Accomplishments that we're proud of

  • Split Decision Engine — A single AI call generates both cooking paths (hack vs. shop) with safety-scored substitutions
  • Five Gemini APIs chained in one user flow with zero external dependencies
  • AI image editing in a cooking app — Users can modify recipe illustrations with natural language prompts mid-cooking
  • 100% free tier — No paid APIs, no external services

What we learned

  • AI Studio's Vibe Coding makes complex multi-API apps accessible to non-coders
  • Concise, copy-paste code blocks work better than lengthy descriptive prompts in AI Studio
  • User-facing terminology matters more than technical accuracy — "Can I trust this swap?" beats "Cross-contamination risk assessment"

What's next for Rescue Chef AI

  • Voice-first cooking — Full conversational flow using Gemini Live API
  • Pantry memory — Persistent ingredient tracking across sessions
  • Community cookbook — Share and remix rescued recipes
  • Waste impact dashboard — Aggregate savings over time with environmental metrics

Built With

Share this project:

Updates