How it worksFor schoolsEvidenceAboutSign inRequest a demo →
How it works

AI grades the lesson.
Mathematics proves the score.

One recorded lesson goes in. A complete, defensible evaluation comes out, and every step can be checked.

The architecture

Three layers.
One defensible score.

Each layer does one job, and no layer can quietly override another.

01

Intelligence reasons

The AI reads the lesson across languages and scores each strand against cited evidence. It never touches the math, and it never hears the raw audio.

02

Computation verifies

A statistical audit sits between reasoning and publication. It runs on code, not language models, and measures how sound each score is before anyone sees it.

03

Educators set the standard

Experts calibrate the rubric. Anything the audit doubts goes to a human reviewer. Doubt goes to a person, never silently into a file.

Layer 1 reasons. Layer 2 audits. Layer 3 contextualises.

The audit gate

The AI judges.
Mathematics checks its work.

An AI can sound just as confident when it's inventing as when it's right, so we never take its word for it.

the audit gate
reasoningverified ✓
CATCHES THE GUESS

Consistency

Every judgement is read many times over; a score that wobbles is held, never published.

CATCHES THE INVENTION

Grounding

Every claim must trace to a real, locatable moment in the audio, checked by signal processing that cannot itself hallucinate.

CATCHES THE DRIFT

Baseline

Each score is weighed against the teacher's own pattern, so anything unusual surfaces for a human.

We're open about what's live today and what's still being built, and we say which is which. Honesty is part of the product.

What happens to a lesson

Passes the audit? Published.
In doubt? A human sees it first.

Published

A lesson that clears every applicable check becomes a verified report, evidence and all.

Quality review

Anything the audit is unsure about is routed to review before it goes near a teacher's file.

Human review

A score the audit doubts is held for a human educator. The decision stays with a person.

The report

A complete report,
not a lonely number.

Behind every score is a rubric forged over thousands of real classroom observations, then made to prove itself, moment by moment.

Timestamped quotes

Every finding cites the exact moment it came from, in [00:17] form, with the excerpt.

آ

Language analysis

Xyova tracks switching between Urdu, English and Arabic, including frequency and tone, mid-sentence.

S1

Speaker analysis

Speaking-time split, students anonymised as "Student 1 / Student 2". Zero names, ever.

Longitudinal profile

Sessions over time, a personal baseline, moving average and "Personal Best" labels.

Verified, in every report

Each report carries proof it passed an independent layer of statistical checks.

A human gold standard

Experts calibrate the standard and review anything in doubt. It builds a benchmark that can't be bought, only earned.

Built for classrooms that switch
language mid-sentence.

English, Urdu and Arabic in a single breath. Single language tools mishear these lessons. Xyova was multilingual from the first line of code.

See it for yourself

Bring one recorded lesson.
See the proof.

Request a demo and evaluate a lesson of your own.