How-to
·
·
⏱ 2 min read
·
👁 10
GPS + face attendance for teachers — the honest version
How to verify a teacher arrived at school without buying a biometric machine, what GPS plus face scan really proves, and the privacy limits you must respect under DPDP.
T
Technokrits
Technokrits team
Most "AI face attendance" pitches make grand claims. Here is what actually works on a normal Android phone, in a 50-student rural school, with no extra hardware:
The teacher opens `/attendance/self` on their phone. Two things happen:
1. The browser shares the device GPS coordinates. The server checks they're within the geofence you defined (default 50m) for the school. If they aren't on campus, attendance is rejected.
2. A 128-number "face vector" — never the photo — is computed in the browser via a lightweight TensorFlow.js model. It is compared against the vector enrolled at first sign-in. Match → attendance recorded. No-match → rejected with a friendly retry button.
This costs nothing per day, no API fees, and a typical model load is 6MB once (cached after).
Privacy bits worth taking seriously:
- The face *vector* is one-way and unique to this app. It cannot be reused to identify the teacher elsewhere.
- The photo never leaves the device.
- Teachers see exactly what data is stored. You can wipe an embedding with one click.
What this DOESN'T prove: that the teacher actually entered a classroom. They could mark themselves at the gate and go home. Pair this with class-wise student attendance (which the same teacher marks 30 minutes later) for the real signal.
The teacher opens `/attendance/self` on their phone. Two things happen:
1. The browser shares the device GPS coordinates. The server checks they're within the geofence you defined (default 50m) for the school. If they aren't on campus, attendance is rejected.
2. A 128-number "face vector" — never the photo — is computed in the browser via a lightweight TensorFlow.js model. It is compared against the vector enrolled at first sign-in. Match → attendance recorded. No-match → rejected with a friendly retry button.
This costs nothing per day, no API fees, and a typical model load is 6MB once (cached after).
Privacy bits worth taking seriously:
- The face *vector* is one-way and unique to this app. It cannot be reused to identify the teacher elsewhere.
- The photo never leaves the device.
- Teachers see exactly what data is stored. You can wipe an embedding with one click.
What this DOESN'T prove: that the teacher actually entered a classroom. They could mark themselves at the gate and go home. Pair this with class-wise student attendance (which the same teacher marks 30 minutes later) for the real signal.
T
Technokrits
Technokrits team — building software for Indian schools.
Try it in your school today