200+ students piloted · 0 audio stored

We don't detect AI. We make it irrelevant.

Structured oral examinations at scale — so understanding is what gets assessed, not writing style.

Book a Demo See How It Works

veridica.ai/session/viva-8f3k

Live session — AI-conducted. Responses recorded. AI does not grade.

Can you walk me through how you approached the morphosyntactic analysis in section two?

I started by identifying the argument structure, then mapped the theta roles across both clauses...

Interesting. What made you choose that framework over X-bar theory?

Student is speaking...

Voice mode — 02:34

Recording

Why assessment is broken

AI detectors are failing the students who need protection most.

20–30%

False positive rate on non-native English speakers

100%

Of TOEFL essays flagged as AI-generated by major tools

AI detectors compliant with EU AI Act (Reg. 2024/1689)

↑

Clear writing has become suspicious

"The problem isn't AI use. It's that we're measuring the wrong thing."

What Veridica does

The viva voce — at scale.

The viva voce is one of the oldest and most reliable forms of academic assessment. A conversation. A student explaining what they know, in their own words, in real time. It's hard to fake. It's even harder to outsource.

Veridica brings this format back — scaled through conversational AI. An AI agent conducts a structured spoken dialogue, probing for understanding, reasoning, and critical thinking. The educator reviews the transcript. The educator issues the grade. Always.

Real-time spoken dialogue

An AI agent conducts a structured oral exam based on educator-defined rubrics, learning outcomes, and the student's own submitted work.

Fully accessible alternative

The same assessment via text. For students with hearing impairments, speech disabilities, social anxiety, or limited bandwidth.

Review, not replace

Full transcripts, audio playback, session flags. The educator makes every decision. No AI-generated grades — ever.

Student	Veridica	Grade
A. Bakker	View transcript →	7.5
M. Konstantinou	View transcript →	_
J. de Vries	View transcript →	8.0

Built differently

Three things that actually matter.

Assessment integrity

AI use becomes irrelevant. Understanding becomes the measure.

When assessment is a conversation, there's nothing to detect. A student who used AI to draft an essay either understands the material — or they don't.

Randomised, artefact-specific questions impossible to script
Tone, hesitation, and spontaneity captured in voice mode
Adaptive follow-up probes reasoning, not just recall

Compliance & fairness

Designed for the EU AI Act. From the first line of code.

Because the AI never issues a grade, it sits outside the high-risk classification under Regulation (EU) 2024/1689. No audio stored. No data used for training.

No audio retention — voice processed in real-time only
Bias monitoring across gender, native language, neurodivergence
DPIA completed; audit logs for every data access event
EU-based hosting — SOC 2 / ISO 27001 certified

EU AI Act

Aligned

GDPR

Compliant

WCAG 2.1

Voice

Real-time

Transcript

Educator

Audio storage AI training data

Educator experience

Works with your LMS. Doesn't replace your judgment.

Standalone web app — no IT project required. For deeper integration, an LTI adapter connects to Moodle, Canvas, Blackboard, eClass, or any standards-compliant LMS.

LTI 1.3 compatible with any standards-compliant LMS
No vendor lock-in — export all data at any time
Role-based access: course-level and institution-level
MFA on all educator accounts

Pilot results

Tested in the classroom. Validated by research.

Piloted across three European countries with students and educators at leading research universities.

200+

Students assessed across English, Spanish, Greek, and Dutch

30+

Educators in pilot design and evaluation

Countries: Netherlands, Spain, Greece

"For the first time, I could see what a student actually understood — not what they could produce with a word processor."

— Pilot educator, University of Twente