A web app that scans your fridge/food inventory using AI and generates recipes based on what you have.

• Frontend: React + Vite
• Auth & Database: Firebase (Authentication, Firestore)
• Python API: FastAPI + Groq (Llama 4 Scout for vision/text, Llama 3.3 70B for recipes)
• Food Photos: Unsplash API (on-demand per recipe)
• Backend Functions: Firebase Cloud Functions (Node.js)

RecipeGenerator/
├── frontend/          # React app
└── backend/
    ├── functions/     # Firebase Cloud Functions
    └── python_api/    # FastAPI image analysis server

• Google and email/password authentication
• Upload a food image → AI detects and lists ingredients
• Review and edit the detected inventory before confirming
• Inventory saved to Firestore per user
• Recipe generation from confirmed inventory
• Recipes saved to Firestore once generated

• Node.js
• Python 3
• Firebase CLI (npm install -g firebase-tools)

cd frontend
npm install
npm run dev

cd backend/python_api
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload

API docs available at http://localhost:8000/docs when running.

cd backend
firebase use recipegenerator-b1315
firebase emulators:start

All endpoints served at http://localhost:8000.

Upload a food image; returns AI-detected inventory. Non-food items are silently ignored.

Request: multipart/form-data with field image (JPEG/PNG)

Response:

{
  "inventory": [
    { "item_name": "Banana", "category": "Produce", "count": 3 }
  ]
}

Parse a free-text ingredient list into structured inventory. Non-food items are silently ignored.

Request:

{ "text": "2 chicken breasts, a dozen eggs, some olive oil" }

Response:

{
  "inventory": [
    { "item_name": "Chicken Breast", "category": "Protein", "count": 2 },
    { "item_name": "Eggs", "category": "Protein", "count": 12 },
    { "item_name": "Olive Oil", "category": "Pantry", "count": 1 }
  ]
}

Takes a confirmed inventory and optional personalisation context; returns 3 recipe suggestions.

Request:

{
  "inventory": [
    { "item_name": "Chicken Breast", "category": "Protein", "count": 2 }
  ],
  "liked_recipes": ["Garlic Chicken Stir-fry"],
  "disliked_recipes": ["Spicy Tofu Bowl"]
}

liked_recipes and disliked_recipes are optional arrays of recipe title strings pulled from the user's saved ratings (≥4 stars / ≤2 stars). The model uses them to personalise suggestions.

Response:

{
  "recipes": [
    {
      "title": "Garlic Chicken Stir-fry",
      "description": "A quick and flavourful weeknight chicken dish.",
      "ingredients": ["Chicken Breast", "Garlic", "Soy Sauce"],
      "steps": ["Slice chicken into strips.", "Heat oil in pan.", "Stir-fry 8 minutes."],
      "prep_time": "10 minutes",
      "cook_time": "15 minutes",
      "servings": 2
    }
  ]
}

Fetches a landscape food photo from Unsplash for a given recipe title. Called on demand when a user opens a recipe detail view.

Query param: title — the recipe title string (URL-encoded)

Response:

{ "image_url": "https://images.unsplash.com/..." }

Returns { "image_url": null } if no matching photo is found.

users/{uid}
  ├── inventory: [...] (array field — full ingredient stock)
  └── recipes/{recipeId}
        title, description, ingredients, steps,
        prep_time, cook_time, servings,
        isFavorited, rating, feedback, savedAt

Security rules ensure each user can only read/write their own documents.