professional - Sole Developer
Helmet Scan-to-Fitment Pipeline
A deterministic pipeline that turns 3D head and helmet scans into a per-station padding prescription.
- Role
- 3D Scanning & Fitment Software Intern
- Team / Org
- LIGHT Helmets · sole developer
- Timeline
- May 2026 – Aug 2026
- Skills / Tools
- 3D mesh processing
- Pose search
- Measurement design
- Validation
Overview
As the sole developer during my internship at LIGHT Helmets, I built a seven-stage pipeline that takes a 3D head scan and a helmet-interior scan and produces a per-station pad-and-shim prescription, end to end for two helmet families and five shell sizes. I designed it to be auditable: measurements are byte-reproducible, and validation gates refuse to emit output when a check fails.
What I did
Built all seven stages: scan cleanup and mesh repair, a canonical coordinate frame derived from geometry, texture-based station labeling, a 4-D seated-pose search, per-station gap measurement, and automated output on the company’s fitment form.
Replaced a coordinate-descent pose solver that stalled at its first iteration with a derivative-free grid search over 52,052 seating poses, cutting head-shell interference from 141 penetrating points to 0.
Designed a deterministic, byte-reproducible measurement instrument and validation gates that block unverified output across a 4,500-file project tree.
Built an Android head-scanning app in Kotlin using ARCore’s Depth API to capture head geometry for the pipeline.
Outcomes
- 141 → 0
- Penetrating points
- 52,052
- Seating poses searched
- 7
- Pipeline stages