EklipseVerifi
Try the demo

Verification over detection

Know what yourstudents canexplain.

Eklipse Verifi adds a short, submission-specific conversation to essays and discussion posts, then links each explanation to the claims, sources and rubric criteria a person needs in order to decide. The decision is theirs, recorded under their name.

Try a 90-second verification See the faculty evidence

Faculty decide. Verifi organises the evidence.

Oversight attestation

BIOL 240 · Discussion 4

Period 3–7 March · 28 learners · 4 decisions shown

Sample
  1. 9f3a
    Dr. A. Okafor09:14
    AI suggestion: Approved · consideredUnderstanding confirmed
  2. c71e
    Dr. A. Okafor09:31
    AI suggestion: Rejected · consideredFollow-up required
  3. 2b8d
    Prof. L. Mensah11:02
    AI suggestion: Edited · consideredUnderstanding confirmed
  4. e4c0
    Prof. L. Mensah11:20
    AI suggestion: Ignored · consideredUnable to determine

Assessment lead

Signature · date

chain head
e4c0…91b7

What the assessment lead signs

One page, per cohort, per period.

Every row is a person: who decided, when, what they did with the suggestion the model offered, and the hash that ties the entry to the one before it. No score appears on it, and no part of it is written by a model.

For your DPO

Read the trust pack

Try it — no signup

Paste a paragraph. See what Verifi would ask.

Verifi reads the text, maps the claims, and shows the exact question a student would answer, and how the answer becomes evidence for a person to review.

Peer reply from Maya: But doesn't high redundancy also waste energy the system could use to grow faster in stable conditions?

Claim map — pick one to verify

The question Verifi would ask

You wrote: “I think biodiversity strengthens ecosystem resilience because species respond differently to disturbance, so the syst…” — walk me through why you reached that rather than an alternative explanation.

Adaptive follow-up: “Maya replied that redundancy wastes energy in stable conditions. How would you answer her?”

Choose a sample student answer

Choose an answer to see the linked evidence.

Sample data, illustrative output. Verifi does not judge authorship, and nothing you type here is retained.

The broken assumption

A polished submission is no longer proof of understanding.

A student can paste a prompt into a generative tool, submit the result, and never read it. Detection tools guess authorship and misfire on the students least able to defend themselves. Neither tells you what the student actually understands.

Detection asks: did a machine write this?
Probabilistic. Biased against non-native writers. Unappealable.
Verification asks: can the student explain it?
Evidence-based. Fair across response modes. A person decides.

Not “did AI write this?” but “can the student explain the specific claims they made?”

The question Verifi is built to answer

The discussion-board wedge

Keep the discussion. Verify the understanding.

Other tools replace the discussion board with a spoken exam or a video ticket. Verifi keeps the written LMS thread exactly as it is and adds a short, claim-specific verification to it.

Without Verifi

  1. 01Instructor posts the discussion prompt
  2. 02Student pastes the prompt into a generative tool
  3. 03A polished post and two replies appear
  4. 04Instructor has no evidence of comprehension

With Verifi

  1. The written post stays in the LMS thread
  2. Verifi selects one specific claim and a peer connection
  3. The student answers a 90-second verification
  4. Faculty get linked evidence, and decide

Adopt it without redesigning the course. Measurement, not policing.

The evidence chain

One chain. Every conclusion inspectable.

The gold line is not decoration; it is the product. Every observation traces both directions, from a rubric criterion back to the exact spoken words and the claim in the submission.

Rendered from the actual Verifi faculty workbench — no mockups, no stock footage.
  1. 01

    Submission

    The student's exact written claim

    “…species respond differently to disturbance…”

  2. 02

    Claim

    One specific statement, mapped to a source

    Linked: Holling 1973.pdf

  3. 03

    Question

    Grounded in that claim, never a trick

    “Why does redundancy speed recovery?”

  4. 04

    Explanation

    Spoken or typed, in the student's words

    0:48 · transcript aligned

  5. 05

    Evidence

    Observation citing the exact excerpt

    Names the mechanism, not a restatement

  6. 06

    Decision

    A named person's outcome of record

    Understanding confirmed · Dr. A. Okafor

The student experience

Students know exactly what happens.

What is recorded, why it is used, and when it is deleted, disclosed before anything starts. The interviewer is named as an AI. Every path has an equal alternative.

Preflight disclosure
Modality, data, time, retention: shown up front, in plain words.
Private practice
Rehearse without faculty ever seeing it.
90-second conversation
One anchor question, one adaptive follow-up.
Low-bandwidth audio
The normal mode. Video is never required.
Typed accommodation
A fully equal path, not a fallback.
A confirmation, not a result
The student sees that their evidence was submitted, never a score.

The faculty experience

Create, review, decide, learn.

The evidence workbench is the product, not a decorative report.

Build a verification

Paste instructions. Approve the plan.

Import an assignment or discussion from your LMS, or paste the prompt. Verifi drafts a claim map and a short question plan grounded in the rubric. You edit every question before it goes live.

  • AI-use policy is an assignment setting, not fine print
  • One anchor question and one adaptive follow-up by default
  • Audio, typed and faculty-led are equal response modes

Verification plan · draft

1Explain why redundancy speeds the recovery pathAnchor
2Respond to Maya's energy-cost replyFollow-up
Time budget90 seconds

In one sentence

Verifi is evidence of what a student can explain. It is not a detector, a judge, or a score.

Verifi is

  • Evidence of explanation and application
  • Submission-specific
  • Faculty-controlled
  • Transparent about data
  • Accessible across response modes

Verifi is not

  • An AI-writing detector
  • A generic chatbot
  • An automatic misconduct judge
  • Covert proctoring
  • A confidence, gaze or accent scorer

Gold is the evidence path Blue is an action or a system state Green appears only when a person has decided

Research and validation

Method, not a badge.

Verifi is built on the Multi-Modal Verification Framework (MMVF). We publish the method and the validation plan, and label every claim with its evidence state rather than calling ourselves “research-backed”.

The MMVF methodFramework
Five modalities: temporal, linguistic, interactive, visual, contextual. Verifying comprehension, not authorship.
Validation planPlanned
Agreement with blinded faculty review, inter-rater reliability, false-escalation and missed-gap rates, fairness across accents and languages.
Model limitationsPublished
Where the system should not be used, and why misconduct decisions stay outside automated evaluation. Instructions for use and Annex IV technical documentation.

A peer-reviewed validity study is in progress. No accuracy figure is published before it completes.

Trust and accessibility

Written for a procurement review.

Every state is labelled precisely. We never write “certified” when we are planning an audit, and the same list drives the sign-in page, so no two surfaces can disagree.

Read the trust packAccessibility statement
EU AI ActIn progress
Treated as a high-risk use in education. Human oversight is by design: every decision of record is a named person's, and the AI's suggestions are recorded with what that person did with them. Exportable oversight records and the Annex IV technical file are in progress.
GDPR / UK GDPRIn progress
Data processing agreement on Article 28 terms is in preparation for pilots. The privacy notice covers what is recorded, the lawful basis, retention and learner rights.
Data residencyIn progress
Single-region hosting today; the region and every subprocessor are named in the trust pack. Customer-selectable residency is not offered yet.
DPIA supportIn progress
A DPIA input pack (processing description, data flows, retention schedule, human-oversight design) is being assembled for pilot institutions.
SubprocessorsIn progress
The list of subprocessors (hosting, media storage, transcription and question generation) is published in the trust pack and changes are notified in advance.
FERPA postureIn place
Student-consented context only; nothing is scraped from the LMS beyond the assignment being verified.
Data ownershipIn place
The institution owns its data and can export or erase it.
Model-training policyIn place
Student data is never used to train models.
Retention and deletionIn place
Per-activity recording windows, a scheduled purge, per-learner export and audited erasure. The window is an institution setting.
Audit and human controlIn place
Faculty decisions cannot be edited after they are recorded; deletion happens only through audited institutional erasure. Full provenance for every AI suggestion, and an exportable per-session oversight record.
EncryptionIn progress
In transit everywhere. At rest: requested on every media object and backup; verification of the host volume is pending, so it is stated as in progress until it is.
Accessibility conformance reportIn progress
WCAG 2.2 AA is the target; typed, audio and faculty-led paths are equal. An accessibility conformance report is being prepared.
LTI 1.3In progress
Blackboard and Canvas launch, roster (NRPS) and completion passback (AGS) are built; a round-trip inside a customer LMS is still pending.
Institutional SSO / SAMLPlanned
Not built yet. OIDC federation (Entra, Okta, Google; Shibboleth via a bridge) is the next identity milestone and replaces the closed demonstration gate before any institutional rollout. No self-service accounts.
Multi-factor authenticationPlanned
Not built yet. TOTP on the demonstration gate is scheduled ahead of federation.
SOC 2Planned
Type II audit is planned and not yet certified.

Integrations

Honest about what is live.

Available

In place
  • —Secure share links
  • —CSV roster import
  • —Evidence export
  • —Private cloud media storage (S3-compatible)
  • —Automated transcription and question generation

Built, validation pending

In progress
  • —Blackboard and Canvas LTI 1.3 launch
  • —Roster sync (NRPS)
  • —Completion passback (AGS)

No completed round-trip inside a customer LMS yet.

Planned

Planned
  • —SSO / SAML
  • —SOC 2 Type II audit

No LMS logo is shown until that integration is live and documented.

Start small

Verify one assignment before changing your whole strategy.

One course, one section, one term. You collect the evidence; you decide whether it earns a wider rollout.

Duration
One term, scoped in writing before it starts.
Included
Setup, faculty onboarding, the evidence workbench, and a full evidence export at the end.
You supply
One assignment or discussion prompt, a faculty lead, and a roster or LMS course.
Data handling
Your institution owns the data. Recordings follow the retention window you set; erasure on request.
Exit
No automatic renewal. Export and deletion confirmation on request.

Pricing posture

Pricing is stated in writing before a pilot starts. It is never tied to outcomes, referrals or the number of students reviewed, because there is nothing here to flag. Ask for the pilot terms at [email protected].

Built for whoever carries the risk: faculty verifying one assignment, departments calibrating outcomes, institutions setting policy, retention, audit and LTI.

Request a one-assignment pilot

Duration
One assignment, one course, one term: typically four to six weeks from the first student session to the faculty debrief.
What is included
Set-up of the verification with the instructor, the student journey (spoken, typed or faculty-led), the faculty evidence workbench, a debrief with the evidence organised for the committee, and a named contact throughout.
What the institution supplies
An instructor and one assignment, student notice through the course, and a contact for IT or the data protection office. LMS launch through LTI 1.3 is available; a secure link works without it.
Data handling
The institution owns its data. Recordings expire on the window the instructor sets; transcripts, decisions and the audit trail are exportable and erasable on request. A data processing agreement on Article 28 terms is offered before any student session.
Exit terms
No fee for the pilot and no obligation after it. At the end the institution decides; on request, its data is exported to it and erased, and the erasure is logged.
A person replies within two working days from [email protected].

Your name, email, institution and role are stored to answer this request and for nothing else. See the privacy notice.