GCP: BigQuery · Looker

🧬 Behavioral AI Engine

Every candidate has a behavioral fingerprint. Deviations from that fingerprint trigger instant scoring and alerts, the most powerful anti-cheating signal that can't be faked.

Live Anomaly Detection

Response Time Anomaly

Risk: High

User normally reads 2.3s/question → now averaging 0.4s

Response time 5.75x faster than established baseline. Possible answer-sheet use.

BASELINE

2.3 s/question

CURRENT

0.4 s/question

Mouse Movement Pattern Shift

Risk: Medium

New trajectory clusters detected: 3 previously unseen movement archetypes

Cheat Pattern Cluster Match

Risk: Critical

Answer latency matches known cheat pattern cluster #47

BigQuery Schema: behavioral_events

CREATE TABLE behavioral_events (
  session_id       STRING NOT NULL,
  candidate_id     STRING NOT NULL,
  event_type       STRING,         -- 'gaze' | 'keystroke' | 'mouse' | 'response_time'
  timestamp        TIMESTAMP,
  value_numeric    FLOAT64,
  value_json       JSON,
  baseline_value   FLOAT64,
  deviation_score  FLOAT64,
  risk_contribution FLOAT64,
  flagged          BOOL DEFAULT FALSE,
  cluster_id       STRING,         -- matched cheat pattern cluster
  INDEX (session_id),
  INDEX (candidate_id, timestamp)
);

Behavioral Risk Score

82

0: Clean100: Critical

Score: 82/100- Human escalation triggered

Response Time: Baseline vs Current

Baseline

Current

How Behavioral Fingerprinting Works

1Pre-exam baseline collected during orientation (5 min)

2BigQuery stores 200+ behavioral signals per session

3Real-time deviation scoring every 500ms

4ML model trained on 2.4M prior sessions identifies cheat patterns

5Risk score feeds directly into Decision Engine