04 / Local applications
AYQYN
Observe with context.
Leave decisions to people.
A Windows local proctoring prototype: participant consent, camera preparation, on-device observations and a separate human review of events.
Codex was used to prepare the code, interface and project materials. YOLO and MediaPipe models were used as provided; the team did not train them.
01
Getting started
Prepare a local session with participant consent and camera calibration. A test scenario and browser replay let you explore the workflow without a camera.
02
The result
An observation timeline, reviewer decisions with reasons, and a JSON report without images.
03 / Case study
How it works
YOLO and MediaPipe process frames on the device in a separate process. Processing and snapshot consent are separate; saving snapshots is off by default.
The reviewer sees duration, face count, frame quality and model confidence for phone detection. They confirm or dismiss the observation and explain their decision.
JSON reports contain no images. The public browser replay shows saved detector results as diagrams without running a camera; photographs are used in the local test replay.
04 / Case study
Project limits
A prototype for Qostanai AI Industry Hackathon 2026, Case 3. Gaze direction is approximate; observations do not prove misconduct and require human review. This is not a certified examination environment. Browser replay does not test camera accuracy.
Project documentation