APPLICANT — Online Examination Platform
A secure online examination platform supporting candidate management, automated assessment workflows, monitoring, and examination administration.
Enterprise · 2023 — 2024 · Lead Full-Stack Engineer
Technology
- ASP.NET Core
- C#
- React
- OpenCV
- SQL Server
- SignalR
- WebRTC
- Identity
Results
- 50K+ Exams Proctored
- 99.4% Identity Match
Overview
Remote exams only work if cheating is actually hard, not just against the rules. APPLICANT gives universities and certification bodies a way to administer exams remotely while still catching the things that matter: someone else in the room, eyes leaving the screen, a tab switch mid-assessment.
That monitoring runs on computer-vision anomaly detection over live WebRTC video — multi-face presence, gaze tracking, tab-switch flags — sitting on top of timed assessment state machines built with ASP.NET Core and SQL Server.
The assessment state machine has to be strict about what counts as a valid session: a dropped WebRTC connection mid-exam isn't automatically treated as an abandoned attempt, but it is flagged for proctor review, and the timer doesn't quietly keep running against a candidate who lost connectivity through no fault of their own.
The challenge
Problem: Naive tab-switch and face-detection flagging produces a flood of false positives — a candidate glancing at a second monitor, a brief network hiccup dropping a video frame, or someone adjusting their webcam angle all look identical to genuine cheating signals if you just count raw events.
Approach: Moved from single-event flagging to a scored anomaly model: individual signals (gaze diversion, face absence, tab switch) accumulate a weighted score over a rolling window instead of triggering an immediate incident, and only sustained or repeated patterns escalate to a flagged incident a human proctor actually reviews.
Result: Incident review queues went from mostly-noise to mostly-actionable, so proctors spend their time on real anomalies instead of triaging false alarms — and the audit trail (snapshot + score history per incident) gives institutions something defensible when a flagged result is disputed.
Key features
- Computer-vision-assisted identity verification during exam sessions
- Real-time anomaly detection (multiple faces, face absence, gaze diversion, tab switching)
- Weighted anomaly scoring over a rolling window to reduce false-positive incident flags
- WebRTC live stream telemetry with automated snapshot incident logging
- Randomized question banks with section-locking timers and clipboard security
- Automated exam grading with performance analytics and compliance audit reports
- Live proctor oversight dashboard with SignalR real-time alerts