01 / 09, 2025

NutriCam

AI food scanning that reveals nutrition, ingredients, and hidden components from a meal photo.

Role

Team engineer, iOS, Flask backend, AI nutrition workflow

Stack

SwiftUI · Combine · Charts · PhotosUI · Flask · Google Gemini · MongoDB · JWT

The Problem

Nutrition tracking breaks when users have to manually search foods, estimate portions, and enter every ingredient. NutriCam's README frames the opposite workflow: point the camera, capture a meal, and let AI identify the food.

The harder product requirement is not only visible food recognition. The app also needs to infer hidden ingredients such as oils, spices, marinades, and cooking components, then turn that analysis into editable nutrition data users can trust.

The Architecture

01SwiftUI iOS app

The mobile client is built with SwiftUI, Combine, native Charts, PhotosUI camera access, URLSession networking, a dark interface, dashboards, weekly insights, streaks, and personalized nutrition goals.

02Flask + Gemini analysis backend

A Python Flask API receives compressed meal images, calls Google Gemini for food and ingredient analysis, calculates nutrition, and returns editable meal results to the app.

03MongoDB and JWT account layer

MongoDB stores user and nutrition data, while JWT authentication secures requests and keeps the dashboard, diary, goals, and account flows tied to the right user.

Decisions that mattered

1.

Expose hidden ingredients

The README emphasizes hidden ingredients as a core differentiator, so the case study now treats hidden oils, spices, and marinades as central analysis output rather than a side note.

2.

Keep results editable

AI nutrition estimates are useful only if users can correct them. Editable ingredients and instant nutrition recalculation turn the model output into a practical food diary entry.

3.

Make tracking more than scanning

The product includes calories, water, exercise, weight, weekly charts, streaks, and goals, so the camera workflow feeds a broader habit-building dashboard instead of ending at a one time analysis screen.

The Numbers

iOS 17+

native app target

Gemini

vision analysis

JWT

secure auth

MongoDB

nutrition storage

What it taught me

The user value is not just recognition; it is the full loop from photo to editable meal to long term dashboard history.

For AI health workflows, correction UX matters as much as model confidence because nutrition data becomes personal recordkeeping.