Home
Learn
Certification
Framework
Summit
Community
Careers & Research
Face Detection verified candidate identity in 0.3s2s ago
EXAM INTEGRITY INTELLIGENCE

AI-Powered Exam Integrity

Next-generation AI proctoring that ensures exam integrity through real-time computer vision, behavioral analysis, and intelligent monitoring — all running client-side in the browser with zero installs.

Zero-Install Browser-Based Privacy-First Client-Side Real-Time Computer Vision
LIVE MONITORING FEED
Face Detection
verified candidate identity in 0.3s
2s ago
Gaze Tracker
confirmed on-screen focus for 12 minutes
5s ago
Browser Lockdown
blocked tab switch attempt
9s ago
Integrity Engine
maintained 98.4% confidence score
14s ago
Warning Manager
escalated from advisory to caution
19s ago
Face Matching
confirmed 99.1% identity match against enrollment
24s ago
8-SYSTEM ARCHITECTURE

Comprehensive exam monitoring

Eight interconnected systems work together to provide layered exam integrity — from computer vision input to intelligent enforcement and real-time feedback.

👁
SYSTEM 01

Face Detection

MediaPipe-powered real-time face detection with temporal smoothing and multi-frame validation for robust presence monitoring

MediaPipe Integration478 LandmarksTemporal SmoothingMulti-Frame ValidationPresence DetectionOcclusion Handling
🎯
SYSTEM 02

Gaze Estimation

Iris landmark tracking for precise screen focus monitoring with configurable tolerance zones and off-screen detection

Iris TrackingScreen Focus MappingOff-Screen DetectionTolerance ZonesAttention HeatmapsCalibration Engine
🔐
SYSTEM 03

Face Matching

Deep learning identity verification comparing live video feed against enrollment photo with anti-spoofing measures

Enrollment MatchingCosine SimilarityAnti-SpoofingLiveness DetectionConfidence ScoringPhoto Comparison
📊
SYSTEM 04

Integrity Scoring

Multi-signal fusion engine computing real-time exam integrity confidence from all monitoring subsystems

Signal FusionConfidence ScoreWeighted AggregationThreshold ManagementHistorical BaselineAnomaly Detection
🛡
SYSTEM 05

Browser Lockdown

Comprehensive browser security preventing tab switches, copy/paste, dev tools access, and fullscreen exit

Tab Switch DetectionFullscreen EnforcementDev Tools BlockingCopy/Paste PreventionRight-Click DisableKeyboard Shortcut Guard
SYSTEM 06

Warning Manager

Intelligent violation escalation system with weighted scoring and progressive alert levels from advisory to critical

Progressive AlertsWeighted ViolationsEscalation TiersAdvisory LevelCaution LevelCritical Threshold
🤖
SYSTEM 07

AI Proctor Agent

Intelligent monitoring agent with contextual decision-making that distinguishes genuine violations from normal behavior

Behavioral AnalysisContext AwarenessPattern RecognitionFalse Positive ReductionDecision EngineReal-Time Inference
🖥
SYSTEM 08

Proctoring Overlay

Real-time visual feedback layer displaying status indicators, warnings, and confidence metrics to the candidate

Status IndicatorsWarning DisplayConfidence MeterCamera PreviewConnection StatusTimer Integration
BY THE NUMBERS

Precision at every layer

0
Face Landmarks Tracked
<0ms
Detection Latency
0
Monitoring Systems
0%.2
Accuracy Rate
PROCTORING PIPELINE

From video capture to integrity decision

Every frame passes through a five-stage pipeline that transforms raw video into actionable integrity intelligence in under 200 milliseconds.

1
INPUT

Video Capture

Camera stream at 30fps with WebRTC optimization

Camera Feed
Enrollment Photo
Browser Events
2
DETECTION

Computer Vision

MediaPipe face mesh and iris landmark extraction

Face Detection
Gaze Estimation
Face Matching
3
ANALYSIS

Signal Fusion

Multi-signal integrity scoring with weighted aggregation

Integrity Score
Behavioral Analysis
Anomaly Detection
4
ENFORCEMENT

Action Layer

Progressive warning escalation and browser lockdown

Warning Manager
Browser Lockdown
Proctor Agent
5
OUTPUT

Feedback Loop

Real-time overlay and exam integrity reporting

Proctoring Overlay
Score Report
Audit Trail
VIDEO FRAME → FACE MESH → GAZE VECTOR → IDENTITY MATCH → INTEGRITY SCORE → ENFORCEMENT → OVERLAY
INTEGRITY CLASSIFICATION

Four-tier confidence scoring

The Integrity Engine continuously computes a confidence score fusing signals from all monitoring systems. Each tier triggers appropriate responses.

EXCELLENT
95-100%

Candidate fully engaged with consistent gaze, verified identity, and no violations detected.

Face present
Gaze on-screen
Identity confirmed
No tab switches
GOOD
80-94%

Minor deviations within normal behavior tolerance. Brief off-screen glances or momentary face occlusion.

Occasional glances
Brief occlusion
Within tolerance
Advisory issued
CAUTION
60-79%

Multiple soft violations accumulated. Extended off-screen activity or repeated tab switch attempts.

Multiple violations
Extended off-screen
Tab attempts
Caution warning
CRITICAL
0-59%

Severe integrity compromise detected. Prolonged face absence, identity mismatch, or repeated critical violations.

Face absent
Identity concern
Critical violations
Exam flagged
PROGRESSIVE ESCALATION

Weighted warning system

The Warning Manager uses exponential weighting to distinguish between accidental deviations and deliberate violations, ensuring fair and proportional responses.

1
ADVISORY
1x

Informational notice — brief gaze deviation or momentary face partial occlusion

2
CAUTION
2x

Moderate concern — extended off-screen activity or repeated soft violations

3
WARNING
4x

Serious concern — tab switch detected, face absent for extended period

4
CRITICAL
8x

Severe violation — identity mismatch, dev tools opened, or multiple critical flags

Advisory
Caution
Warning
Critical
WHY AI PROCTOR

Built different by design

Privacy-first architecture with zero-install deployment. All computer vision runs client-side — no video data ever leaves the candidate device.

Zero-Install

Runs entirely in the browser with WebRTC — no downloads, plugins, or extensions required

Privacy-First

All computer vision processing happens client-side — no video streams leave the candidate device

Adaptive Thresholds

AI-calibrated sensitivity that adjusts to lighting, camera quality, and candidate behavior

Accessibility-Aware

Accommodations for screen readers, alternative inputs, and reduced-motion preferences

Real-Time Scoring

Continuous integrity confidence score updated every frame with multi-signal fusion

Progressive Warnings

Four-tier escalation from advisory to critical with weighted violation scoring

Anti-Spoofing

Liveness detection prevents photo and video replay attacks during identity verification

UNDER THE HOOD

Technical specifications

COMPUTER VISION
Face Mesh ModelMediaPipe 478-landmark
Iris Tracking10 iris landmarks
Detection Rate30 fps real-time
SmoothingTemporal + spatial
Anti-SpoofingLiveness detection
OcclusionPartial face handling
INTEGRITY ENGINE
Scoring ModelMulti-signal fusion
Update FrequencyPer-frame (33ms)
Signal Inputs6 subsystems
Threshold Tiers4 confidence levels
BaselineRolling 60s window
Accuracy99.2% validated
BROWSER SECURITY
Tab DetectionVisibility API
FullscreenEnforced + monitored
Dev ToolsDetection + blocking
ClipboardCopy/paste disabled
Context MenuRight-click blocked
ShortcutsKey combo guarded
CANDIDATE EXPERIENCE

Seamless exam journey

From enrollment to certification, the proctoring system works invisibly in the background — candidates focus on the exam, not the technology.

01

Enrollment

Candidate captures enrollment photo for identity baseline

02

System Check

Camera, browser, and connectivity verification

03

Identity Match

Live face verified against enrollment photo

04

Active Monitoring

Continuous AI proctoring throughout exam session

05

Integrity Report

Final score and session audit trail generated

SIGNAL INTELLIGENCE

Multi-signal fusion for accurate decisions

No single signal determines integrity. The system fuses multiple data streams with weighted confidence to reduce false positives and ensure fairness.

Face Presence

25%

Continuous face detection with temporal smoothing. Momentary occlusions (sneezing, scratching) are tolerated. Extended absence triggers escalation.

Detection confidence: 0.0 - 1.0

Gaze Direction

20%

Iris landmark tracking maps gaze to screen regions. Short off-screen glances are normal. Sustained off-screen attention is flagged.

On-screen ratio: 0% - 100%

Identity Match

20%

Live face compared to enrollment photo using deep feature extraction. Cosine similarity threshold with liveness verification.

Similarity score: 0.0 - 1.0

Browser Integrity

15%

Monitors tab visibility, fullscreen state, dev tools status, and keyboard shortcuts. Binary violation events with immediate detection.

Violation count: 0 - N

Behavioral Pattern

10%

AI agent analyzes overall behavioral patterns. Distinguishes natural exam behavior from suspicious activity using contextual reasoning.

Anomaly score: 0.0 - 1.0

Temporal Consistency

10%

Evaluates signal stability over rolling time windows. Sudden changes in multiple signals simultaneously indicate potential compromise.

Stability index: 0.0 - 1.0
PRIVACY ARCHITECTURE

Your data stays on your device

AI Proctor is engineered with privacy as a foundational principle. All computer vision and behavioral analysis runs entirely in the browser.

Client-Side Processing

MediaPipe WASM runs in-browser
No video frames transmitted
Camera feed stays local
WebRTC for capture only

Data Minimization

Only scores are reported
No biometric storage
Session data auto-expires
Enrollment photos encrypted

Transparency

Real-time status overlay
Visible confidence score
Warning explanations
Pre-exam disclosure

Compliance

GDPR-aware design
Accessibility standards
Accommodation support
Audit trail for review
EXPERIENCE AI PROCTORING

Exam integrity you can trust

Take a proctored exam and see the AI monitoring system in action. Real-time computer vision, intelligent analysis, and instant feedback — all running privately in your browser.

Zero-install browser-based Privacy-first client-side AI Real-time integrity scoring