Kalorientracker
ExperimentalNative iOS app that estimates the calories and macros of a meal from a single photo. The user points the camera at their food, the image is analyzed by a vision AI, and the result is logged into a daily diary with progress rings, macro breakdown charts and weekly statistics. A barcode scanner and a German product database back the manual entry path.
github.com/SchmidieDE/kalorientrackerProblem
Manual calorie tracking is tedious: users have to look up every food, weigh portions and type in macros, so most people abandon their food diary within days.
Solution
A SwiftUI app where a meal photo is sent to a vision model (Gemini cloud mode) that returns calories, protein, carbs and fat. Entries are stored per meal category and visualized with a daily calorie ring and macro charts. A Node/Express backend serves fuzzy product search, barcode (EAN) lookup against a Postgres product database, and an analyze proxy, with Sign in with Apple and account sync on top.
Tech Stack
Infrastructure
iOS client built with XcodeGen (project.yml, Sign in with Apple entitlement). Backend is a Dockerized Node.js/Express service behind Traefik at api.kalorientracker.webgantic.com, backed by a shared Postgres (kalorientracker DB) with pg_trgm fuzzy search and rate limiting; auth/sync via self-hosted Supabase at supabase-kalorientracker.webgantic.com.
Outlook
On-device inference is currently a stub (planned llama.cpp + Qwen vision GGUF integration for offline analysis). Next steps are completing the local model path so analysis works without network, expanding the product database, and a possible App Store release.