Heal

AI-powered diabetes nutrition assistant that Snap & analyze your meals in seconds

Comment

Inspiration

Managing diabetes requires constant attention to nutrition—counting carbs, tracking macros, and making split-second meal decisions. Traditional food tracking apps require tedious manual entry, making them impractical for daily use. I wanted to create something effortless: just snap a photo and let AI do the rest.

What it does

Heal is an AI-powered nutrition assistant specifically designed for people with diabetes:

  1. Personalized Profile: During registration, the app calculates your daily calorie and macro targets using the Mifflin-St Jeor equation based on your height, weight, age, activity level, and diabetes type (T1D/T2D).

  2. Photo-Based Tracking: Take a photo of your meal, and GPT-4o vision analyzes it to identify foods, estimate portions, and calculate macros (protein, carbs, fat, calories).

  3. Real-Time Feedback: Instantly compare your meal against per-meal and daily targets, with visual progress bars showing how you're tracking.

  4. Actionable Suggestions: GPT-4o-mini generates specific, implementable tips like "reduce rice by 50g" or "add 100g steamed vegetables."

  5. Daily Insights: At the end of each day, receive a summary of your eating patterns and 3-4 focus points for tomorrow.

The goal: zero manual tracking, maximum health impact.

UI/UX Excellence Highlights

Apple-Native Design Language

  • Dynamic Type Support: Full integration with iOS Dynamic Type—users can adjust font sizes from 12pt to 34pt+ for perfect readability
  • Automatic Dark Mode: Seamless system-wide dark/light mode switching that adapts to user preferences and time of day
  • SF Symbols Integration: Native Apple iconography for consistent, familiar visual language
  • Haptic Feedback: Subtle vibrations that provide tactile confirmation for important actions

One-Tap Photo Analysis

  • Seamless Camera Integration: Full-screen camera experience with instant photo capture
  • Smart Visual Feedback: Real-time macro breakdown with beautiful, easy-to-read progress bars
  • Intuitive Gestures: Swipe to dismiss, tap to retake—natural mobile interactions

Intelligent Data Visualization

  • Color-Coded Progress Bars: Green (on track), yellow (approaching limit), red (exceeded)
  • Adaptive Dark Mode: Perfect contrast and readability in any lighting condition
  • Micro-Interactions: Smooth animations that provide immediate visual feedback

AI-Powered Smart Suggestions

  • Contextual Recommendations: "Add 1/2 cup of vegetables" instead of generic advice
  • Visual Portion Guidance: "Remove 2 slices of bread" with clear visual indicators
  • Personalized Insights: Tailored to individual diabetes type (T1D vs T2D)

How we built it

Architecture

Backend (Python/FastAPI):

  • Modular design with separation of concerns (routes → handlers → services)
  • OpenAI GPT-4o for vision-based food analysis
  • OpenAI GPT-4o-mini for text generation (comparisons, suggestions, summaries)
  • Structured JSON outputs using OpenAI's schema validation
  • Pydantic for type-safe request/response validation

iOS App (Swift/SwiftUI):

  • Native camera integration for seamless photo capture
  • MVVM-inspired architecture with @Published reactive state
  • URLSession for async networking
  • Custom JSON decoders with CodingKeys for Python ↔ Swift compatibility
  • Dark mode support and adaptive UI

Challenges we ran into

  1. iOS Multipart Upload: FastAPI wasn't receiving images from iOS's URLSession. Solved by debugging with test endpoints and manually parsing request.form() to identify field names.

  2. JSON Decoding Mismatches: Python's snake_case vs Swift's camelCase required custom CodingKeys for every struct. One missing mapping caused silent failures.

Accomplishments that we're proud of

End-to-End AI Integration: Successfully integrated GPT-4o vision for food analysis and GPT-4o-mini for intelligent suggestions, creating a truly automated nutrition tracking experience.

Robust Error Handling: Debugged and solved complex iOS ↔ FastAPI multipart upload issues, JSON decoding mismatches, and OpenAI schema validation errors.

Production-Ready Architecture: Clean modular backend (7 files) and iOS app (11 files) with proper separation of concerns, making it easy to maintain and extend.

Dark Mode Native: Properly implemented adaptive colors and layouts that work seamlessly in both light and dark modes.

Real Device Testing: Got the app working on actual iPhone with camera, network debugging, and real-world photo analysis—not just simulator testing.

Built With

Share this project:

Updates

Submission history