FlavorFlick - An AI Recipe App

flavorflick.ca | Swipe. Learn. Eat.

An AI-powered recipe discovery app that learns your category-specifiic unique taste preferences through swipes and delivers personalized meal recommendations.

Core Features

• 5,593 Recipes - Across 5 meal categories with full nutrition data and images
• Implicit Feedback Learning - Learns preferences from swipes
• Taste-Based Recommendation Engine - Using Euclidean distance similarity in 5D taste vector space to recommend recipes
• Category-Aware Learning - Separate taste profiles per meal type
• 80/20 Exploration/Exploitation - Balancing to prevent filter bubbles
• Filtering - Filter by diet type and meal type while scrolling
• Tinder-Style Swipe Interface - With physics-based animations

---

Tech Stack

Backend

• FastAPI + Python - 2 REST endpoints: /recipes/feed for recommendations, /swipes/ for recording interactions
• SQLAlchemy - Async database queries, defines Recipe/User/Swipe models with relationships and cascade deletes
• Pydantic - Validates incoming swipe data and serializes recipe responses

Database

• PostgreSQL - 3 tables: recipes (5,593 rows with JSONB taste profiles), users (device-based identification), swipes (stores like/dislike with taste profile snapshots)
• Alembic - Database migrations for schema changes

Frontend

• React Native + Expo + TypeScript - 3 screens (Home, Liked, Settings)
• PanResponder + Animated API - Detects swipe gestures and animates card rotation/translation

Infrastructure

• AWS EC2 - Hosts FastAPI server at api.flavorflick.ca
• Supabase - Managed PostgreSQL with connection pooling
• Cloudflare - DNS routing, SSL certificates, CDN for images
• Expo EAS - Builds iOS/Android apps for distribution

---

Architecture

┌──────────────────┐      HTTPS       ┌──────────────────┐      TCP        ┌──────────────────┐
│  React Native    │  ←───────────→   │  FastAPI         │  ←──────────→   │  PostgreSQL      │
│  App (Expo)      │                  │  (AWS EC2)       │                 │  (Supabase)      │
├──────────────────┤                  ├──────────────────┤                 ├──────────────────┤
│ • Swipe UI       │                  │ • /recipes/feed  │                 │ • recipes        │
│ • Context State  │                  │ • /swipes/       │                 │ • users          │
│ • Device ID      │                  │ • Rec Engine     │                 │ • swipes         │
└──────────────────┘                  └──────────────────┘                 └──────────────────┘
        ↑
        │ CDN
┌──────────────────┐
│   Cloudflare     │
│ • SSL/DNS        │
│ • Image Caching  │
└──────────────────┘

Flow: User swipes → Backend records interaction → Rebuilds taste profile from history → Returns similarity-sorted recommendations

---

Recommendation Algorithm

Content-based filtering system that learns user taste preferences through swipe interactions.

• 5D Taste Vectors - Recipes mapped to 5 taste categories
• Euclidean Distance Scoring - Produces 0-1 match scores for recipes
• Category-Aware Profiles - Seperate profiles for each mealtype
• Sorted by Similarity - Recipes returned in descending order by match score using SQL JOIN between recipes and user profiles
• 80/20 Exploration/Exploitation - Balances personalization with discovery to prevent filter bubbles
• Negative Feedback Overcorrection - Dislikes weighted at 30% to prevent a few bad swipes from dominating

---

Data

Recipe Dataset

• 5,593 recipes with pre-computed taste profiles, nutrition data, ingredients, and step-by-step instructions

Swipes Table

• Table with all user swipes: user_id, recipe_id, direction
• Foreign keys to users and recipes tables with cascade deletes for referential integrity

User Data

• Device-based identification - No authentication required, UUID generated on first launch
• Category-specific profiles rebuilt on each feed request from swipe history

---

Frontend

• Swipe Interface - PanResponder + Animated API for gesture detection and physics-based card animations
• State Management - 3 React Context providers (Feed, Preferences, Liked) for global state
• 3 Screens - Home (swipe cards), Liked (saved recipes grid), Settings (filters)

Deployment

Backend (AWS EC2)

• Instance - t3.micro running Ubuntu, hosts FastAPI via Uvicorn
• Domain - api.flavorflick.ca routed through Cloudflare

Database (Supabase)

• PostgreSQL 15 - Managed instance with automatic backups
• Connection Pooling - PgBouncer for efficient connection management with async SQLAlchemy
• Migrations - Alembic tracks schema changes with version control

CDN & DNS (Cloudflare)

• SSL/TLS - Full strict encryption between client, Cloudflare, and origin
• DNS - A records for API, CNAME for main domain
• Caching - Static assets and recipe images cached at edge

---

Project Structure

recipe-app/
├── backend/
│   └── app/
│       ├── main.py          # FastAPI entry point
│       ├── models.py        # SQLAlchemy ORM
│       ├── services/        # Recommendation engine
│       ├── routes/          # API endpoints
│       ├── crud/            # Database operations
│       └── schemas/         # Pydantic models
│
├── frontend/
│   └── src/
│       ├── screens/         # Home, Liked, Settings
│       ├── components/      # Reusable UI components
│       ├── context/         # React Context providers
│       └── api/             # API client
│
└── website-frontend/        # Marketing website (flavorflick.ca)