← Crafts
Muster preview

Muster

Biometrics pilot Full-stack engineer

Proof-of-concept facial recognition attendance for classrooms — camera capture matched against enrolled student embeddings.

Problem

Manual roll call burns instructional minutes and is easy to game. Muster prototypes camera-based check-in: enroll a reference face set, match live frames, and write attendance rows a teacher can audit.

What it covers

  • Webcam capture pipeline with face detection pre-filter
  • Embedding match against enrolled classroom roster
  • React UI for session start, overrides, and audit export
  • Confidence thresholds with manual fallback for ambiguous matches
frame + session id lookup embeddings student id + confidence write attendance row Classroom camera Match service Roster store
Attendance match flow
export async function recordAttendance(sessionId: string, frame: ImageBitmap) {
  const faces = await detectFaces(frame);
  if (faces.length !== 1) return { status: 'ambiguous' as const };

  const match = await matchEmbedding(sessionId, faces[0].embedding);
  if (match.confidence < 0.82) return { status: 'low_confidence' as const };

  await db.attendance.insert({ sessionId, studentId: match.studentId, at: Date.now() });
  return { status: 'recorded' as const, studentId: match.studentId };
}
Muster attendance UI placeholder
Screenshot placeholder — swap for the classroom check-in screen.