AI-proof assessment

Verify
understanding
in the AI era.

After every submission, Viva holds a short voice conversation with each student about their own work. Your school gets verifiable evidence of learning, spoken in the student's own words.

Schools: pilot this term. Instructors and students can start free — no payment info needed.

Verified by Viva
VC
Via Cho
Renewable Energy Proposal · 5m 42s
CONSISTENT
UNDERSTANDINGHigh
AUTHENTICITYHigh

The student owns the core argument and defended her data choices under follow-up questions. Explanations matched the submission throughout.

Voiceprint matched · ID V-8F3A
LIVE CONVERSATION · 03:41

“You chose a t-test in section 3. Why that over ANOVA for this data?”

0%instructor-rated report accuracy
~0 minper interview, zero marking time
0independent AIs cross-check flags
0%of final calls made by instructors
Try it yourself

Talk to Viva.

Step into a student's shoes for a moment, then see the report you get back.

Viva · Sophie Live
Talk to Viva Tap to start, then just speak. (Or type.)

A scripted taste. On the live product, Viva adapts with AI, in the student's language, after every submission.

The report you get back

The depth is in the report.

Every interview becomes a report tied to what the student actually submitted: what they understood, where they were shaky, and proof it was really them. Watch it read itself.

Jordan Reyes · Newsvendor problem set defense Verified
Transcript · on their own submission
You stocked 460 units when mean demand was 500. Walk me through why you ordered below the mean.
JRBecause here the overage cost is higher than the underage cost, so the critical ratio lands below 0.5Explained the newsvendor tradeoff in their own words, and it matches the method in the submitted essay. and you pull the order under the mean.
Good. And if the two costs flipped?
JRThen the ratio goes above 0.5 and you'd stock above the mean. I'm less sure on the exact cutoff though.
ConsistentOwns the core tradeoff
UnderstandingHigh
Newsvendor tradeoffSolid
Critical ratioSolid
Sensitivity to cutoffShaky
AuthenticityVoiceprint 98%
Same speaker as the enrollment baseline for the full session. No mid-interview voice swap detected.
Independent AI cross-check3 / 3 agree
GPT-4o concurs: reasoning is the student's own
Claude concurs
Gemini concurs
Every student. Every submission. This depth, automatically.
The problem no one has named

Everyone knows students use AI.
No one can tell who actually learned.

Instructors feel it every day. The work comes in polished, and they can no longer tell what the student actually knows. Detectors false-flag honest students and miss lightly-edited AI. Banning it is unenforceable. Rewriting every assignment is an arms race you lose. So the grade goes out, and no one, instructor or student, is sure learning happened.

86%

of students already use AI in their studies. The question was never whether they use it. It's whether anyone can still tell what they learned.

50%+

false-positive rates for AI detectors on non-native English writing. Guessing from text punishes honest students and proves nothing.

The gap

Every school feels this. Almost none can name it precisely, or respond in a way that actually works. Naming it is where the fix begins.

How it works

Three steps to verified work

STEP 1

Students submit anything

Essays, lab reports, code, slides, posters, spreadsheets. Nothing changes about how you assign.

Drop any file: PDF, DOCX, images, code, XLSX
essay.pdflab_report.docxanalysis.xlsxposter.pngmodel.py
STEP 2

Viva talks with each student

A 5 to 10 minute voice conversation about their work, in their language. Adaptive follow-ups probe the reasoning behind every claim. Identity is confirmed by voiceprint.

CONVERSATION · 04:12
Your poster argues Coptis outperforms two antibiotics. What would make that comparison unfair?
If the concentrations weren’t matched by daily intake… which is why we normalized to 4 to 6 grams…
Good — so you normalized to match real intake. If you hadn't, which way would that have biased the result?
STEP 3

Instructors get a verified report

A clear verdict with evidence: understanding and authenticity scores, quotes tied to the document, integrity signals, and a second opinion from independent AIs. You stay the judge.

JM
Jake M.
Gatsby essay · 6m 05s
FLAGGED

Could not explain the thesis he submitted. Vocabulary in conversation didn’t match the essay’s. Independent AIs corroborate. Suggested: give a second chance →

…and leaders see it across every class

Verification rates, flags, and follow-ups in one view. Department heads know where learning is proven, and where to look closer, without chasing spreadsheets.

Harvard · Fall 2026
Overview
Assignments
Review
Students
Insights
All classes / Fall 2026
Expos 20: Ethics & AI
24 students · 3 essays · 41 interviews completed
Things to do
2 interviews flagged for review in the close-reading essayReview →
5 students haven’t started the research essay defenseRemind →
AssignmentDueTakenVerifiedStatus
Research essay: oral defense
Submitted work · 10 min
Jul 1219 / 2479%Open
Close reading: argument & evidence
Submitted work · 8 min
Jul 524 / 24100%2 flagged
Rhetorical analysis defense
Submitted work · adaptive
Jun 2823 / 2496%Closed
Built for trust

Evidence, not accusations

Every verdict is backed by layered, inspectable signals. Schools adopt Viva because the report holds up when a parent asks why.

🎙️

Voice identity

A short enrolled voiceprint confirms the same student is speaking in every session. Consent-based, biometric-grade.

🛡️

Independent cross-check

AIs from different providers re-score flagged interviews blind. Agreement becomes a trust badge on the report.

🖼️

Visual probes

Generated diagrams test real understanding on the spot. Answers read from a second screen don’t hold up to them.

📄

On-document evidence

Report claims link to highlighted passages in the actual submission, PDFs, docs, even scanned posters via OCR.

👀

Focus & video signals

Optional camera and tab-focus checks for high-stakes exams. Off by default, instructor-controlled per assignment.

⚖️

Instructors stay the judge

Viva never makes the final call. It surfaces evidence, suggests second chances, and records your decision.

Zero new workflow

Works with what your school already uses

Single sign-on launch and grade passback through LTI 1.3. Rosters sync nightly from your SIS. Students never manage another account.

LTI 1.3OneRosterGoogle ClassroomSAML SSOCSVGrade passback
Student data, guarded

Data security is our priority

Encrypted in transit and at rest. Recordings auto-delete on a 30-day clock. Voiceprints are consent-based and never used for anything but identity. A signed DPA for every school.

Strict data-residency requirements? Viva Private runs single-tenant inside your institution's own cloud.

FERPACompliant
COPPACompliant
K-PIPACompliant
GDPR
UK/EU
Compliant
Viva Credential

Proof students can carry with them.

Every verified conversation can issue a Viva Card: a tamper-evident credential a student attaches to an application, portfolio, or transcript. Anyone can verify it. Only the school can revoke it.

  • Tamper-evident. Each card is sealed with a SHA-256 hash. Change one detail and verification fails.
  • Verifiable by anyone. A scannable link resolves to Viva, with no login required.
  • Owned by the school. You issue, you revoke. Credentials stay under institutional control.
Verified
Maya Sullivan
Expos 20: Ethics & AI · Harvard · Jun 2026
UnderstandingHigh
AuthenticityConsistent
Verification IDVIVA·8F3A…C1D9

Well-grounded, insightful questions about the student's own work, designed to evaluate what they actually understand. It's like a really good verbal defense with every assignment — very hard to fake or cheat.

Dr. David Dixon · Head of AI Education & Innovation, MIT

A school’s promise is simple: its graduates earned what they know. Viva is the first tool I’ve seen that lets an institution keep that promise in the age of AI, with proof instead of suspicion.

Mr. Ted Hill · President, Roessler-Chadwick Foundation

Trust is the one thing a school cannot outsource. With Viva, every grade we give carries a conversation behind it, and when a parent asks how we know, we can show them.

Mr. Euiseob Lee · Chairman, Ivy Collegiate School
FAQ

Questions schools ask

Can it really tell whether work is a student’s own?+

Viva doesn’t guess from text like detectors do. It asks the student to explain, defend, and extend their own submission out loud, with adaptive follow-ups no script can prepare someone for. Inconsistency shows up as evidence, not a probability score.

What about shy students, accents, or second languages?+

Conversations run in the student’s language (English and Korean today, more coming), grading never punishes accent or hesitation, and instructors can set equality accommodations, including a human-run 1:1 alternative for any student.

How much instructor time does this take?+

About zero per student. Interviews run themselves after submission; you review a one-screen report, and most classes need a closer look at only 2 to 3 flagged results. Grades pass back to your LMS automatically.

What if a student gets outside help during the conversation?+

Layered safeguards make live coaching impractical: voiceprint identity, natural-pace conversation, generated visual probes, optional focus and camera signals, and cross-checking by independent AI models.

What does it cost?+

Instructors and students can start free. Schools license per student per year, with pilot pricing that includes every premium feature. Ask us for the one-page pricing sheet.

Bring proof of learning to your school.

Pilot in one department this term. Set up takes an afternoon, not a semester.