Mobile apps
Brickster
An iOS app that uses the camera, AR, and on-device ML to find the LEGO pieces for a set and track which ones you've collected.
Goals
- Pick a LEGO set and fetch its parts list
- Point the iPhone at a pile of bricks; the app detects pieces and highlights the ones you need in AR
- Track which pieces you’ve picked up
Design
Native SwiftUI with no third-party dependencies. Recognition is a two-stage model:
- A single-class
lego_pieceYOLO11n detector, exported to CoreML with NMS at 640px - A MobileNetV3-Small classifier for the mold, limited to the parts in the chosen set
Color is analyzed in the app. Training data is synthetic: LDraw parts rendered in Blender, with Brickognize used as a labelling oracle.
Status
The brickster-ml pipeline CLI works end to end through rendering and dataset assembly (17 tests
green). Training and checks on real photos are next.
Related
- BricksterRN: a React Native take using zero-shot MobileCLIP instead of a classifier trained per set
Changelog
- 2026-06-10: App analysis and ML architecture brainstorm
- 2026-07-05: ML pipeline repo built