MeatBrains
Reads meat-case shelf tags and ranks packs by value.
What it is
MeatBrains reads supermarket meat-case shelf tags through the iPhone camera. From each tag it pulls the price per pound, weight, total, cut, USDA grade and sell-by date, then ranks the packs in view by value, explains the grade premium, flags markdowns and links recipes for the cut. The app was built over 101 commits on 2026-09-02 and 09-03; later QA and fix cycles ran through 2026-09-19.
By the numbers
What it means for your team
It shows coding agents working in parallel lanes against written tickets and test gates, and an AI feature held back because it missed an accuracy bar set before the work started.
Status
Checked on 2026-10-08: the backend's 261 tests and the 134 tests in the two core Swift packages pass. MeatBrains has run only in the simulator: never on a physical iPhone, no TestFlight build and no users. The product direction was left open on 2026-09-18. The recipe clips come from a video library that would need creator licences before public use.
On-device reading
Apple Vision reads the text on each tag. A value is accepted only if price times weight equals the total within $0.02 and at least three reads agree.
Parallel lanes
The app is split into five Swift packages: tag parsing, scanning, value math, interface and data upload. Four lanes, three for the app and one for the backend, worked in parallel in separate git worktrees and merged without conflicts.
Backend
A FastAPI service stores anonymous scan data, serves recipes from the Maillard library and lets a user delete their data.
Testing without a camera
The iOS Simulator has no camera, so recorded tag moments from store walkthrough videos are replayed through the real pipeline instead.
An accuracy bar set first
The marbling feature had to reach 80% correct, with at most 20% abstention, on 85 or more real pairs. Free data supported only a 26-pair test, which scored 19 of 26, or 73%, so the feature was withheld.