Smart Fridge

Turn fridge photos into planet-friendly meals—AI that identifies your ingredients, measures their carbon impact, and helps you cook and shop more sustainably

Comment

Inspiration

We were inspired by the enormous problem of global food waste.
According to the UN, about 1.3 billion tons of food are wasted each year, creating nearly 8–10% of global greenhouse gas emissions. We wanted to build something that helps people reduce food waste at home and make sustainable food choices easier.

What it does

Smart Fridge is a web application that:

  • Accepts multiple photos of a fridge or pantry.
  • Uses AWS Rekognition to detect and normalize ingredients into a clean inventory list.
  • Prompts for the number of people you’re cooking for and uses an AI agent (OpenAI GPT-4o) to generate recipes with the ingredients you already have.
  • Scores each ingredient’s carbon footprint using a curated dataset and highlights high-impact foods like beef.
  • Suggests low-carbon swaps and calculates a household sustainability score to guide future shopping decisions.

How we built it

Before coding, we created a logical flow chart to map the process:
Image Upload → Ingredient Detection → Inventory Merge → Carbon Scoring → AI Recipe Generation.

Tech Stack:

  • Frontend: Next.js + Tailwind for a responsive UI with multi-image upload and results display.
  • Backend: FastAPI for REST endpoints, orchestration, and data validation.
  • Image Analysis: AWS Rekognition for food detection and RapidFuzz for label normalization.
  • AI Agent: OpenAI GPT-4o mini for recipe generation and sustainability suggestions.
  • Data: Custom JSON mapping ingredients to their CO₂e (kg per 100 g) and sustainability tags.
  • Deployment: Frontend on Vercel, Backend on Render.

We divided the work into four key areas:
Frontend UI, Vision Service (Rekognition + normalization), Planner & Carbon Engine, and Recipe Service (LLM integration with strict JSON schema).

Challenges we ran into

  • AI Tuning: Ensuring the AI only generated recipes using the detected ingredients required strict JSON schemas and careful prompt engineering.
  • Merge Conflicts: With four beginners collaborating, Git merge conflicts were frequent and required quick resolution.
  • Feature Scope: We debated which features to prioritize, balancing time constraints with our sustainability focus.
  • API Limits & Latency: Managing Rekognition calls and OpenAI token limits while keeping response times short.

Accomplishments that we're proud of

  • Building a fully working end-to-end system in under 24 hours.
  • Designing a clean architecture that integrates computer vision, carbon scoring, and an AI recipe generator.
  • Creating a simple but effective carbon impact scoring system with actionable low-carbon swap suggestions.
  • Learning how to collaborate efficiently with Git despite frequent conflicts.

What we learned

  • How to design and deploy a multi-service architecture combining real-time image analysis and AI.
  • Effective Git collaboration with branching, pull requests, and conflict resolution.
  • Prompt engineering and how to enforce structured JSON output from large language models.
  • The importance of clear, impactful sustainability metrics that motivate users to act.

What's next for Smart Fridge

  • Automatic Expiration Detection: Use computer vision to estimate expiration dates and alert users.
  • Personal Sustainability Goals: Let users set targets (e.g., reduce carbon footprint by X% over time).
  • Community Features: Integrate with local food donation or composting resources to help people keep food out of landfills.
  • Enhanced Recipe Engine: Add filters for nutrition, allergy preferences, or regional cuisines.

Built With

Share this project:

Updates