Here’s the problem
The work has to make sense before the interface can.
An edge that looks plausible on camera is not automatically a safe cut path. Calibration, geometry, and operator authority all matter.
Computer vision + manufacturing automation · 2026
PatchSight MK2
PatchSight tackles the production side of the PATCH22 ecosystem. The native Mac app now brings together OAK-D Lite preview, exterior run-stitch detection, calibration, and operator-reviewed SVG export. The target is the white stitch line around the patch—not every bright detail inside it.
- My role
- Product and systems engineering
- Status
- Stationary-bed MVP
- Contribution
- Vision pipeline, calibration rules, operator review, and safe export boundaries.
- Python
- OpenCV
- DepthAI
- SwiftUI
- Shapely
What made this difficult
The environment is part of the product.
- Camera and calibration failures must remain visible.
- Open, invalid, or out-of-bounds geometry must fail closed.
- The system cannot control the laser; an operator approves and exports the file.
Decisions that shaped it
I made the hard boundaries visible.
Trace the stitch, not the artwork
The detection pipeline targets the white exterior run stitch and proposes separate contours for review. Internal white artwork should not become a cut path.
Make review unavoidable
The detected contour is shown over the source image before an operator accepts or rejects it.
Keep machine authority out
The MVP exports calibrated SVG geometry for manual import instead of driving the laser directly.
What I built
Enough of the system to test the real seam.
- Python 3.12 engine using DepthAI, OpenCV, Shapely, NumPy, and Pydantic.
- A SwiftUI operator app with live preview, run-stitch tracing, overlay review, and explicit approval before export.
- Homography calibration, invalid-path rejection, immutable job records, and SVG dimensions in millimeters.
What is true now ·
The stationary-bed MVP now has a native operator workflow from preview to reviewed SVG export. Physical scale and cut quality are still the next proof points.
What still needs proving
- The MVP assumes a stationary work bed and a configured, calibrated camera.
- Physical tolerances, Beam Studio import scale, and repeatable test cuts still need acceptance. The app does not upload jobs or start the laser.
Next
Agree on physical tolerances, verify SVG scale in Beam Studio, and measure test cuts across representative patch shapes.