AI proctoring built into the LMS — not bolted on.
Most LMS vendors point a third-party iframe at a proctoring vendor. When the iframe fails at 9 AM, nobody owns the incident. Cognaxa runs proctoring as a first-class subsystem — face presence, audio anomaly detection, server-synced timer, confidence bands, and a mandatory human review queue. One SLA, one support channel, one vendor on the line.
Six things native proctoring gets right that bolted-on iframes don't.
Native — not a third-party iframe
The proctoring engine runs inside the same product as the quiz engine, the gradebook, and the audit log. One SLA. One on-call. One vendor on the line at 9 AM on exam day.
Face presence + identity check
Per-session face detection and presence sampling. Live identity check against an institution-supplied roster photo, with explicit consent capture and a per-tenant retention policy.
Audio + secondary-device anomaly detection
Background voice detection, off-screen device cues, and shared-screen heuristics. Every flag is timestamped against the exam timeline, not the wall clock.
Server-synced exam timer
The clock is authoritative server-side. Tab switches, network drops, and clock-tampering attempts cannot grant extra time. The browser is the renderer; the server is the source of truth.
Reviewer queue with confidence bands
Every flagged session lands in a reviewer queue with confidence bands. Auto-pass the obvious clean sessions, surface the genuinely ambiguous ones to a human reviewer in a single screen.
Audit-ready evidence pack
Every proctored attempt produces a downloadable evidence bundle: timeline, flags, reviewer notes, and consent record. Plays back as a single linked timeline for academic-integrity hearings.
AI-assisted, not AI-deciding.
Education buyers ask about accuracy, bias, auditability, and human review. We do not claim "human-level accuracy" on a marketing page — we publish how the system actually works, what it captures, and where it surfaces uncertainty.
Confidence bands, not binary verdicts
Every AI signal carries a confidence score. Reviewers see "uncertain" instead of false certainty, and institutions set their own pass/review thresholds per assessment.
Human review queue is mandatory
Cognaxa does not auto-fail or auto-flag students from AI alone. Every disputed flag is reviewed by a human before a result is final.
Bias documentation in the artefact pack
Our AI grading whitepaper documents the evaluation pipeline, training-data provenance, and known failure modes. Available to procurement under NDA.
Accommodations are first-class
Per-learner extra time, alternate format, and reduced-monitoring profiles are part of the assessment config — not an afterthought patched in by the proctor.
Procurement-ready
Questions academic-integrity teams ask.
Is this AI-only proctoring, or is there a human in the loop?
AI-assisted with mandatory human review. Cognaxa surfaces flagged sessions to a reviewer queue with confidence bands. The institution sets the thresholds; we never auto-fail a student on AI signal alone.
How does Cognaxa handle accessibility and accommodations?
Accommodations are configured per-learner at the assessment level: extra time, alternate question formats, reduced-monitoring profiles for students with documented needs. The proctoring engine respects every accommodation flag set in the gradebook.
What about student privacy?
Cognaxa captures only the data the institution configures (face presence, audio anomaly, browser focus events). Recordings are tenant-scoped, retention is policy-controlled, and consent is captured per attempt. Our DPA + subprocessor list are at /trust/downloads.
Can it scale to peak finals load?
Yes. The proctoring engine is a first-class subsystem inside Cognaxa, scaled with the same infrastructure as the quiz engine. Customers run 4,000+ concurrent proctored sessions during finals on Enterprise plans.
See it run on a real exam.
Bring an actual assessment to the demo — we'll proctor it live, walk through the reviewer queue, and show you the audit pack a hearing committee would actually receive.