Exam Center
Book a demo
Assessment & Proctoring

Proctor the exam. Prove the result.

Author, deliver and monitor every exam in one place — and close each case with evidence a candidate can see.

AI flags/a person decides/the candidate can appeal
Question bankAssessmentsSessionsMonitoringReviewAppealsReportsWebhooksTranslationsAI evaluationAudit logRBACQuestion bankAssessmentsSessionsMonitoringReviewAppealsReportsWebhooksTranslationsAI evaluationAudit logRBAC
The console

One screen for every live exam.

Sessions, risk, the review queue, languages, integrations and the audit trail — without opening six tools.

An illustration of the Exam Center dashboardA schematic of the console dashboard covering the whole platform: live sessions, sessions flagged for review, question bank size and credits used; sessions across the day and the risk distribution; the review queue and webhook delivery; and cards for assessments, AI evaluation, reports, roles, integrations, languages and the audit log. This is a drawing, not a screen capture.NORTHVALE INSTITUTE OF TECHNOLOGYLIVEDashboardReports & analyticsCandidatesQuestion bankAssessment librariesAssessmentsSessionsMonitoringReviewGroupsSettingsDashboard · todayLive sessions5in progress now▲ 2 since 9amFlagged2waiting on a person▼ 3% of todayQuestion bank1,284versioned items▲ 18 this weekCredits used1261 per 30 minutes▲ of 2,000Sessions across todayRisk · 62 sessionsLow41Med13High6Crit2Review queue · appeals trackedADCASE-2211CRITICALRICASE-2210MEDIUMIntegrations · webhook deliveryCANDIDATE_SUBMITTED200CANDIDATE_ASSIGNED500 · retriedLanguages · 13HindiliveMarathiliveTamil72%Assessments18 publishedfrozen at publisha sat paper never changesAI evaluationLong answers, scoredwith a written reasonafter submission, not duringReports10 viewsCSV · XLSX · JSONscheduled, or on requestRoles9 rolesdeny by defaultenforced on the serverAudit logEvery significant actionwho · what · whenwith a diff of the change
Interface illustration, not a screen capture. A drawing of the console dashboard — live sessions, the review queue, risk for the day, question bank, languages, reporting, webhooks and the audit log. Figures are one institution's illustrative day, not platform throughput.
How it works

AI flags. A person decides.

Every step to the threshold is automatic. Nothing past it is.

The integrity pipelineA detection produces a violation. The violation contributes weight to a risk score. Past a threshold a review case opens. A person then records the decision — the only step that is not automatic, and one the candidate can appeal.LIVE · ONE SESSIONRISK 62 · HIGH01 · DetectionMODEL OR RULECONFIDENCE 0–100raises02 · ViolationTYPE · SEVERITYRISK WEIGHTweights03 · Risk scoreACCUMULATED WEIGHTTHRESHOLD62 / 50opens04 · Review caseOPENS AT THRESHOLDNO STEP IS AUTOMATIC PAST THE FOURTHto a reviewer05 · A person decidesOne decision, with a reasonAPPROVEWARNINGMANUAL REVIEWINVALIDATEDISQUALIFYappeal reopens
Detection produces evidence, not verdicts. A person records the decision, with a reason — and the candidate can appeal it.
The Platform

Everything an exam needs.

Publish freezes the paperQuestion versions keep moving in the bank. The published assessment holds the version it snapshotted, so later edits never reach a paper someone already sat.Question bankv1v2v3EDITS CONTINUE → v3PublishedSNAPSHOT · v2KEY FROZENA PUBLISHED PAPER IS FIXEDCONTENT · KEY · TRANSLATIONS

Author once

Versioned items, pooled papers, published before use.

The offline outboxWhile the connection is down, answers queue locally. On reconnect they flush in order, and the gap is recorded as a monitoring signal rather than a violation.Queued4 ANSWERS · 2 EVENTSON RECONNECTSERVERIN ORDERMONITORING_GAPNOT A VIOLATIONRESUMES ON THE SAME QUESTION

Deliver anywhere

A link in the browser. Nothing to install.

Six detection lanesSix kinds of signal — identity, gaze, objects, audio, browser activity and connectivity — feed one session timeline. Each detection carries a confidence score.NO_FACEOFF_SCREEN_GAZEPHONE_DETECTEDWHISPERINGTAB_SWITCHMONITORING_GAPONESESSIONEACH WITH A CONFIDENCE SCORE

Watch six ways

Typed detections, each with a confidence score.

Evidence carries its fingerprintEach captured artifact is fingerprinted with a SHA-256 at the moment of capture and shown to the reviewer. Evidence bytes are never exposed by a public link.EvidenceWEBCAM00:14SHA-256SCREEN00:14SHA-256AUDIO00:22SHA-256REVIEWHASHEDSTREAMED UNDER PERMISSION

Decide with proof

Evidence timeline, recorded reason, appeal.

Apply or proposeAI evaluation of a free-text answer either applies its score directly, or proposes it for a human grader to confirm. The mode is set per assessment.FREE-TEXTANSWERAUTO · APPLIEDPROPOSEGRADER CONFIRMSSET PER ASSESSMENTVS A REFERENCE ANSWER

Mark long answers

Free-text and image answers, scored with a rationale.

A signed, replayable webhookSeven triggers, each delivered as a signed, idempotent event with a stable event id, a delivery log, and retry and replay.eventIdtriggerpayloadHMACIDEMPOTENTYOUR API500RETRY7 TRIGGERSEACH SIGNEDASSIGNED · STARTED · SUBMITTED

Push it downstream

Seven signed webhook triggers, retried and replayable.

Monitoring

Six ways we watch.

No layer decides anything on its own. A person always does.

Six monitoring layers, and what happens to what they findOn the left, one live session sits at the centre of six detection layers that watch it at the same time: identity (no face, multiple faces) from the camera, gaze and pose (off-screen gaze, head turning) from the camera, objects (phone detected, book detected) from the camera, audio (whispering) from the microphone, browser activity (tab switch, window focus lost) from the browser, and connectivity (monitoring gap, connectivity lost) from the network. On the right, what those layers produce: typed evidence carrying a confidence score, a weighted risk score whose weights the institution sets per assessment, a threshold that opens a review case rather than reaching a verdict, a person who accepts or dismisses and records one reason, and an appeal tracked to its own outcome.LIVEOne sessionwatched six waysno verdicts01IdentityCAMERANO_FACEMULTIPLE_FACES02Gaze & poseCAMERAOFF_SCREEN_GAZEHEAD_TURNING03ObjectsCAMERAPHONE_DETECTEDBOOK_DETECTED04AudioMICWHISPERING05Browser activityBROWSERTAB_SWITCHWINDOW_FOCUS_LOST06ConnectivityNETWORKMONITORING_GAPCONNECTIVITY_LOSTWATCHINGDECIDING01Typed evidenceevery detection carries a confidence score02Weighted risk scorethe weights are yours, set per assessment03A threshold opens a casea queue item for a reviewer, not a verdict04A person decidesaccept or dismiss, then one recorded reason05The candidate can appealtracked to its own outcome
Every code drawn here is one the platform actually emits.
How it runs

Three steps to exam day.

Building the paperA blueprint draws a fixed number of questions from a pool into a section. The draw happens once, when the paper is authored — not per candidate.PooldrawsSectionDrawn once, at authoring
1

Build the paper

Sections, pools, marking rules.

The availability windowThree availability modes on one timeline. Fixed: one continuous window that every candidate shares, with the same start and the same end. Per candidate: three separate windows, staggered, each the same length. Always open: a window with no closing edge, used for practice and retakes.When it is openFixedPer candidateAlways
2

Open the window

Invite candidates, set the proctoring level.

Closing a review caseA reviewer accepts or dismisses each violation, then records one decision — approve, warning, manual review, invalidate or disqualify. The candidate can appeal.Violations✓✓✕A persondecidesAppealable
3

Close the case

Review the evidence, record the decision.

Translation coverage and the approval gateCoverage is tracked per language and section. AI drafts a translation, but a human must approve it before it can be used, and the approved text is frozen into the assessment at publish — so no AI runs during the exam.CoverageHIMRTABNURAI draftsA human approvesNo AI runs during the exam
13 Languages

One paper. Every language.

Candidates pick theirs when the exam opens — and can switch mid-attempt.

HIहिन्दी
MRमराठी
GUગુજરાતી
TAதமிழ்
TEతెలుగు
KNಕನ್ನಡ
MLമലയാളം
BNবাংলা
PAਪੰਜਾਬੀ
ORଓଡ଼ିଆ
ASঅসমীয়া
URاردو
ENEnglish
Integrations

Connects to what you already run.

Results, violations and decisions leave the platform the moment they happen — signed, idempotent and replayable. No middleware to babysit.

Seven signed triggersAssigned · started · submitted · result published · disqualified · review completed · proctoring
A REST APIScoped by the same roles as the console.
Bulk CSV and XLSXCandidates in, results out, at batch size.
How Exam Center connects to your systemsThree routes out of the platform: signed webhooks with seven subscribable triggers, a REST API, and bulk CSV or XLSX import and export.Exam CenterSessionsResultsViolationsDecisionsWebhooks7 signed triggersREST APIScoped by roleBulk filesCSV · XLSXYourstack
FAQ

Asked first.

No. A one-time link opens the exam in the browser they already have.

Never. Detectors produce evidence with confidence scores. A person reads it and records the decision — and the candidate can appeal.

Answers save as the candidate moves through the paper. A drop is logged as a network event, not a violation, and the session resumes on the same question.

Yes — build them from pools. A blueprint draws from a pool to generate a section, so one question bank produces as many distinct papers as you need. Each published paper is then frozen, and everyone assigned to it sits exactly that paper.

Yes — each is its own tenant, with its own users, libraries, branding and limits, on its own subdomain.

Bring one paper. We'll run it live.

Thirty minutes, your own exam, no slide deck.