AI suggests. The teacher decides.
No grade, no feedback and no exercise reaches a student without a teacher having reviewed and approved it. This page describes precisely the role we give to AI, what we prohibit it from doing, and the safeguards we have put in place to ensure this division of roles is more than a declaration.
Assistance, never substitution
AI prepares work that the teacher then takes over. It closes no pedagogical decision on their behalf.
The teacher's marking scheme is the reference
Marking is based on the marking scheme provided by the teacher, not on the model's general judgement.
Everything is traceable
What was proposed, what was changed, by whom and when: every step is recorded and available for review.
What AI does, and what it never does.
The boundary is not set by a toggle that can be switched off. It is built into the way the product works.
What AI does
- Suggest exercise instructions based on the programme and the teacher's guidelines.
- Read submissions and compare each answer against the marking scheme criteria.
- Propose a grade per criterion, together with the justification supporting it.
- Draft a set of teaching notes for the student.
- Record what it proposed, so that any divergence from the teacher's decision remains visible.
What it never does
- Publish a grade or comment without a teacher's approval.
- Modify the teacher's marking scheme or deviate from it on its own initiative.
- Decide on a student's progression, retention or any form of academic guidance.
- Compare teachers with one another or produce a measure of their activity.
- Use your institution's submissions, marking schemes or exercises to train a model.
Six mechanisms, not six intentions.
Each one is a behaviour of the product, verifiable in the application and documented for your audits.
Mandatory human approval
An assessment remains a proposal until a teacher has approved it. There is no automatic mode, even at an institution's request.
The marking scheme as the sole reference
Marking is done criterion by criterion, against the marking scheme submitted by the teacher. Every point awarded is tied to a criterion, never to an overall impression.
What can be checked without AI is checked without AI
For an exam sat online, numerical answers are marked by direct comparison with the teacher's solution, field by field, without any model involved. The marking scheme and the award rule are declared by the teacher; if absent, the question is passed back to the AI.
Decision log
Initial proposal, teacher's modification, final grade, timestamp: the complete history of an assessment is retained and exportable.
No training on your data
Your institution's submissions, marking schemes and exercises are used solely for your institution. They do not feed any model and are never mixed with those of another institution.
Student's right to contest
An approved grade can be reopened by the teacher, with the reason for the change recorded. The student knows that a teacher assessed their submission.
Where the human sits in the chain
The same sequence applies to all three capabilities: exercise creation, submission marking, teaching notes. Two human checkpoints systematically frame the AI's work.
- Teacher
Submit and frame
Instructions, marking scheme, guidelines, class context.
- AI
Analyse
Reading submissions, comparing each answer criterion by criterion against the marking scheme.
- AI
Propose
Grade per criterion, with the justification supporting it, and a draft set of teaching notes.
- Teacher
Review and decide
They modify, override, rewrite. Nothing is locked.
- Teacher
Publish
It is this approval, and this alone, that makes the grade exist.
What we do not claim to do.
A useful assistant is one whose limits are known. Here are the ones we acknowledge.
AI makes mistakes, and you must be able to see them
A model may misread handwriting, miss a nuance in reasoning, or overmark a well-worded but incorrect answer. That is why the approval screen displays the justification alongside every point awarded: so that any disagreement is visible at a glance.
Pedagogy remains outside our scope
The choice of what is assessed, the level of expectation, what deserves a point or a remediation note belongs to the teacher and the institution. We provide no standard marking scheme, no level framework and no guidance recommendation.
The legal and technical aspects are covered separately
Hosting, encryption, retention periods, sub-processors, AI Act obligations for systems classified as high-risk in education: everything is set out on the dedicated page, in a format your DPO can attach to a compliance file.
The best safeguard is trying it on your own submissions.
Upload an exercise, your marking scheme and a few submissions. You will see exactly what the AI proposes, and what remains for you to decide.