Gavin Wu_

cat ~/projects/nceagrader.md

NCEAGrader

AI grading for New Zealand schools: 22 → 1,435+ users across 10+ schools

visit ↗ ← cd ~ Next.js · Python · MongoDB · AWS
NCEAGrader

NCEAGrader helps New Zealand schools run defensible NCEA-style digital assessments: live proctoring, audit trails, NZQA-aligned rubrics and teacher-reviewed AI-assisted marking. I founded it and I'm the lead engineer.

Growth

From 22 users to 1,435+ (65×) across 10+ schools.

How it grades

  • Rubric-aligned embeddings, so a response is compared against what the marking schedule actually rewards
  • Teacher-calibrated models: teachers' own marks pull the model toward how their department grades
  • Autoscaling inference and embedding caches to survive exam-week spikes
  • Exam proctoring for the authenticity side of the problem

Stack: Next.js, Python, MongoDB, AWS. An OpenClaw instance on an Azure VPS handles email automation and development work.

Policy

I pitched NCEAGrader to the Prime Minister and the Education Minister as a framework for NZQA's digital assessment.