Secure assessments.
Intelligent proctoring.
Trusted results.
EvalenceAI is the infrastructure for creating, delivering, proctoring, evaluating and auditing high-stakes online assessments — with the evidence to defend every result it produces.
- Delivery
- Browser-based, no installer
- Oversight
- AI-assisted, human-decided
- Evidence
- Recorded and reviewable
The lifecycle
One platform for the entire assessment lifecycle.
An exam is authored, sat, supervised, marked and audited in the same system. Nothing is exported to a second tool halfway through and reconciled afterwards, which is where integrity and evidence are usually lost.
- 01
Create exam
Assemble the paper from a tagged question bank, with sections, timing and randomisation.
Exam engine
- 02
Enrol candidates
Bulk import, self-registration or payment-based enrolment, inside a scheduling window.
Admin console
- 03
Verify identity
System check for network, camera and microphone, then identity capture before entry.
Candidate portal
- 04
Deliver assessment
Server-authoritative timing, answer auto-save, focus-loss tracking and resumable sittings.
Candidate portal
- 05
Proctor
Detection raises candidate violations; a proctor watching live confirms or dismisses each one.
Proctor console
- 06
Evaluate
Objective questions auto-mark; written answers route to evaluators against a rubric.
Evaluation engine
- 07
Publish results
Consolidate scores, break them down by section and topic, then release to candidates.
Results
- 08
Audit
The sitting becomes a permanent record: timeline, recording, evidence and sign-off.
Audit trail
Candidate experience
A sitting that fails before it starts is worse than one that never ran.
Most disputed exams are lost to preparation, not cheating — a camera that was never granted, a clock the candidate could reset. The candidate portal closes those gaps before the paper is visible.
- 01
System check before entry
Network, camera, microphone and screen share are verified in sequence. A candidate cannot reach the paper with a setup that will fail mid-exam.
- 02
Identity verified at the door
The candidate is captured once at entry. Every later frame is matched against that capture, so the person who started is the person who finishes.
- 03
Server-authoritative timing
The clock is held by the server. Refreshing, losing the connection or changing device does not buy extra minutes.
- 04
Answers saved as they are given
Each response is persisted on entry and the section shows what has been answered, so a dropped connection costs nothing.
- Network18 Mbps up · stable
- CameraFaceTime HD · 1280×720
- MicrophoneBuilt-in · input detected
- Screen shareEntire screen · required
- IdentityCaptured · pending match
Written response · 12 marks
Q 27 / 40Explain why tenant isolation must be enforced in the data layer rather than the application layer. Refer to at least two failure modes.
Enforcing isolation in the application layer assumes every query is written correctly. A single missing predicate exposes another tenant’s rows…
AI proctoring
AI that assists proctors — not replaces judgment.
Detection narrows thousands of candidate-minutes down to the moments worth looking at. What it never does is decide. Every alert is a candidate violation until a person confirms it, and the dismissal is recorded just as carefully as the confirmation.
- CaptureFrames sampled adaptively
- InferenceVision and audio models
- AggregateDeduplicated, cooled down
- EvidenceFrame, version, confidence
- Human reviewConfirm or dismiss
- Signal
- MULTI_FACE
- Faces found
- 2
- Confidence
- 0.79
- Threshold
- 0.75
- Model
- blazeface · v1.4.0
- Candidate
- S. Koné
- Fairness impact
- −8 (reversible)
Face presence
Candidate absent from frame, or a second person entering it.
Identity match
Live frames compared against the identity captured at entry.
Liveness
Presentation-attack checks against a photo or replayed video.
Object detection
Phones, books, second screens and secondary devices in view.
Gaze and head pose
Sustained off-screen attention, separated from ordinary glances.
Lip movement
Rhythmic speech patterns, with yawns, coughs and smiles rejected.
Audio
Speech in the room and keyword flagging on the transcript.
Focus loss
Tab switching, window changes and ending a required screen share.
Human proctoring
Twenty-five candidates, one proctor, no guesswork about where to look.
The grid carries state, not decoration. A tile changes because something happened, the detail panel explains what, and the controls act on the session immediately.
SK
MH
TG
AB
YN
DW
LA
RB
KM
PF
JT
NE
OZ
CH
VU15 of 25 assigned candidates shown · auto-assigned at one proctor per 25
Sara KonéCAND-40218 · Section BBelow 70 — the session is sampled at a higher frame rate until it recovers.
- 10:26Second person in frame
- 10:14Gaze off-screen — 11s
- 10:02Identity verified
Evaluation
Marking that two people would score the same way.
Objective questions settle themselves. Everything else routes to a marker with the rubric attached and the candidate's name withheld, so the score reflects the answer.
- Automatic marking
- Objective question types score on submission, with per-section totals.
- Evaluator queues
- Written answers are assigned to markers as a queue, not a spreadsheet.
- Rubric scoring
- Criteria carry their own marks, so two markers apply the same scale.
- Blind evaluation
- Candidate identity is withheld from the marker while scoring.
- Consolidation
- Automatic and manual scores merge into one result before release.
- Breakdowns
- Results resolve by section, topic and cohort for reporting.
“Enforcing isolation in the application layer assumes every query is written correctly. A single missing predicate exposes another tenant’s rows, and the failure is silent…”
| Criterion | Awarded | Max |
|---|---|---|
| Identifies the trust boundary | 4 | 4 |
| Names two failure modes | 3 | 4 |
| Justifies the data-layer control | 3 | 4 |
| Total | 10 | 12 |
Audit and evidence
Reconstruct the sitting, months after it ended.
When a result is challenged, an assertion is worth nothing and a record is worth everything. A finished sitting is permanent — the platform never deletes it.
- Click-level activity log for the whole sitting
- Webcam and screen recording, replayable end to end
- Violation bookmarks that jump the player to the moment
- Evidence package exported for an appeal or a regulator
- Legal hold that suspends retention while a result is disputed
- 09:58System check passedCamera, microphone, network
- 10:00Identity verifiedMatch confidence recorded
- 10:01Exam startedServer-authoritative timer begins
- 10:26Object detected — phoneFrame evidence attachedEvidence
- 10:26Proctor reviewedConfirmed · warning issued
- 11:14Face absent — 8sDismissed · candidate adjusted cameraEvidence
- 11:47Exam submittedAnswers sealed
- 12:30Evaluation completedRubric applied, score consolidated
Integrations
Fits the systems the institution already runs.
Assessment is one step in a longer process. Enrolment arrives from somewhere and results have to go back there, which is an API problem before it is a product one.
Enrolment in · results and evidence out
REST API
AvailableExams, candidates, sessions and results.
Scoped API keys
AvailableTenant-bound keys, permissioned per scope.
Webhooks
AvailableSession and result events pushed as they happen.
LTI 1.3
RoadmapCanvas, Moodle and Blackboard launch and grade passback.
SAML 2.0 SSO
RoadmapInstitutional identity provider sign-in.
ATS / HRMS
RoadmapConnectors for enterprise hiring and L&D systems.
Security
Controls in the architecture, not on a badge wall.
High-stakes assessment carries real data about real people, and a platform holding it should be able to say precisely how — not point at a logo.
We do not display compliance certifications we have not been awarded. Where a programme requires a specific attestation, that is a conversation to have during procurement rather than a claim to make on a marketing page.
- Tenant isolation
- Every query is scoped to one tenant, enforced in the data layer with row-level security rather than trusted to the caller.
- Role-based access
- Six roles from candidate to auditor. Each request is checked against a permission before it reads a row.
- Multi-factor authentication
- Available on staff accounts, with sessions that can be revoked centrally.
- Encryption
- TLS in transit and encryption at rest for recordings and evidence.
- Signed URLs
- Recordings are served through short-lived signed links, never a public path.
- Immutable audit records
- Actions are appended with actor, role, tenant and address. A finished sitting is never deleted.
Who uses it
Five roles, one record of what happened.
Each role sees the surface it needs and nothing else — but every action any of them takes lands in the same session record.
- Candidates01
A sitting that works the first time.
- Guided system check before entry
- Identity verification in the browser
- Autosaved answers, resumable session
- Proctors02
Attention directed where it is needed.
- Grid of live candidates with status
- Alerts carrying their own evidence
- Pause, warn, escalate or terminate
- Evaluators03
Marking, without the administration.
- Assigned queues instead of exports
- Rubric criteria with fixed marks
- Blind mode while scoring
- Administrators04
The institution's control surface.
- Exam configuration and scheduling
- Enrolment, branding and users
- Usage, invoices and API keys
- Auditors05
Reconstruct what actually happened.
- Full timeline per sitting
- Recording with violation bookmarks
- Evidence export and legal holds
Pricing
Priced per sitting, not per seat.
Consumption-based, so the bill follows the exams that actually ran. Start without a commitment and move to the annual rate when volume justifies it.
Starter
No annual commitment
Save a card at signup with nothing due today. Billed for the sessions you run.
Get started- Exam engine and question banks
- In-browser proctoring and recording
- Proctor console, staffed by your team
- Automatic and manual marking
- 30-day recording retention
- REST API and scoped keys
Growth
Most chosenFrom 500 sessions a year, prepaid
Commit to an annual volume and the per-session rate halves.
Get started- Everything in Starter
- Half the session rate, prepaid annually
- 90-day recording retention
- Auditor console and evidence export
- Webhooks and extended reporting
- Your branding on the candidate experience
Enterprise
20,000+ sessions a year
For certification bodies, government programmes and procurement requirements.
Talk to sales- Everything in Growth
- Volume pricing and managed proctor staffing
- Extended retention and legal-hold policy
- Data residency and deployment requirements
- Named contact for onboarding and incidents
Prices in USD. A session is one candidate sitting one exam; a retake is a second session. A sitting that never starts — a no-show, or a failed system check — is not billed. Managed proctor staffing is quoted separately.
Questions
What institutions ask first.
If yours is not here, ask us during the demo.
One candidate sitting one exam. A retake is a second session. If a sitting never reaches the in-progress state — the candidate does not appear, or fails the system check — it is not billed.
Nothing. The exam, the detection models and the recording all run in an ordinary browser. Candidates grant camera and screen-share permission at the start of the sitting, and the paper stays hidden until they do.
No. Detection raises a candidate violation with its model version, confidence and frame evidence attached. A proctor confirms or dismisses it during the exam, and an auditor reviews the sitting afterwards. The decision and the person who made it are both recorded.
30 days on Starter and 90 days on Growth, with longer terms by agreement. Access is scoped by role and tenant, recordings are served through short-lived signed links, every view is written to the audit trail, and a legal hold suspends deletion while a result is disputed.
Yes — that is what Starter is for. Nothing is due at signup and you are billed only for sessions you actually run. Move to Growth when your volume justifies the annual rate.
Build assessments your institution can trust.
See the full lifecycle on your own exam material — delivery, proctoring, evaluation and the evidence trail behind a result.