Browser-Native AI Proctoring System

Loom video preferred. We're building a browser-native AI proctoring platform for online exams, technical interviews, and certification testing. No desktop agent, no video upload, everything runs client-side in the browser. Core features needed: - Real-time gaze tracking using MediaPipe face landmarks (iris position, zone classification into center, peripheral, away) - Multi-face detection with sub-second alerting when a second face enters frame - Tab-switch and window-blur detection using the browser Visibility API - Live proctor dashboard showing active sessions and real-time flags via WebSocket - Post-session integrity report with a 0-100 score, gaze distribution, and flag timeline - Candidate-facing exam interface and a separate proctor/admin dashboard Tech stack: - Frontend: MediaPipe (face mesh/iris), WebRTC, WebSocket client - Backend: Go for the real-time WebSocket/session service, Python/FastAPI for any ML-adjacent processing or scoring pipeline - Twilio or equivalent for session infrastructure if needed What to include in your proposal: - Relevant experience with MediaPipe, browser-based computer vision, or proctoring/monitoring systems - Whether you've built real-time WebSocket dashboards before - A rough breakdown of how you'd approach the gaze zone classification and scoring logic - Timeline and milestone-based pricing preference (fixed price or hourly) This is a build from a working reference design, not from scratch conceptually, so ability to move fast and match an existing UX pattern matters more than novel research.