for Work · a division of I-Teach
The pedagogy that works in a classroom is the pedagogy that works at a desk.
I-Teach built its personalization engine inside Israel's national school system, and proved it there. The architecture was never specific to a subject or an age — it encodes how a person moves from the competence they have to the competence they need. We are now bringing it to workforce learning: AI literacy and regulated training, professional upskilling, and employment-integration programmes for adults for whom a lecture hall was never realistic.
Here for schools, networks or the K-12 platform? I-Teach for education01 — The premise
Moving from schools to workforce learning is usually a story about a new market. For us it is a consequence of three things that were true about the engine before there was a market to move into.
i
Education names an institution. Knowledge names a payload to be delivered. Learning is one person moving from the competence they have to the competence they need. The engine encodes that process, not a curriculum — which is why a 40-year-old engineer and a 15-year-old student are the same product question asked of different populations.
ii
The MPC engine does not know mathematics. It knows how to teach. It tracks roughly twenty dimensions of a single learner and adapts inside the session. A new domain is therefore a content-and-partnership question, not an engineering one.
iii
The engine sits behind whatever interface a partner already uses. White-label by design: the partner keeps their brand, their content, their compliance framework and their existing learning platform. We supply the layer nobody can see.
The engine does not know mathematics. It knows how to teach.
02 — Why now
Article 4 of the EU AI Act obliges every organisation operating AI systems to ensure a sufficient level of AI literacy among the staff who use them. It entered into force in February 2025, with enforcement from August 2026. The market's answer so far has been the completion certificate — and a certificate records attendance, not understanding.
Completion is not comprehension
Click-through courses measure whether an employee reached the last slide. A regulator asking whether the workforce understood its obligations is asking a different question, and a completion log cannot answer it.
The certificate is the wrong artefact
What withstands scrutiny is a record of what each person could actually do: explain the concept in their own words, recognise a bad output, decide when a human must stay in the loop. That is an evidence problem, not a content-delivery problem.
One course cannot fit one workforce
A production technician, a finance analyst and a senior developer do not share a starting point. A fixed module either bores the expert or loses the newcomer; a personalization engine gives each of them the version pitched at them.
Mandatory training is disliked, so it is rushed
A dialogue that responds to what a person actually said holds attention in a way a slide deck does not. Engagement here is not a nice-to-have — it is the precondition for the evidence being worth anything.
03 — What we measure
Article 5(1)(f) of the EU AI Act prohibits outright — not merely restricts — AI systems that infer emotions in the workplace. Our measurements were designed on the permitted side of that line, and would be the right measurements even if the line did not exist. What we observe is pedagogic behaviour: what a person did with a problem.
| Observed — pedagogic behaviour | Not observed — prohibited emotional inference |
|---|---|
| How long a task took to resolve | Whether the learner was frustrated |
| The quality of the question the learner asked | The learner's tone of voice |
| How many attempts it took to reach understanding | The learner's level of self-confidence |
| Whether the principle transferred to an unseen context | Performance anxiety |
The distinction matters commercially as well as legally. Transfer to an unseen context is the one signal that cannot be gamed: a learner cannot memorise a scenario they have never met. It is also the signal a completion rate is structurally incapable of producing.
04 — Trust architecture
Learning data about a person is among the most sensitive material an employer can hold. Systems that hand it to managers as a productivity score have been rejected by workforces before, and rightly. Five commitments are built into the product rather than written into a policy.
The individual is the primary data subject of their own learning record — not a by-product of a training programme purchased on their behalf.
Organisation-level and team-level views only. There is no per-employee performance dashboard for managers, because that is a different product and we are not building it.
Declining the personal layer is a supported state inside the system, not an exception that a manager has to approve.
Each employee can see how they learn, where they get stuck and what unblocks them — and decides for themselves whether to bring it into a conversation with their manager.
Not to other clients, not to aggregators, not to third parties. Industry benchmarking, where it happens at all, is anonymised and aggregated.
06 — Two tracks
The same engine, two very different buyers, two very different definitions of success. We keep them distinct rather than blurring them into one pitch.
AI literacy and other regulated training where evidence of understanding is the deliverable; professional upskilling in technical and functional domains; language for the workplace. For distributors, integrators and training providers, the engine is available white-label — you keep the client relationship, the brand and the content; we supply the personalization layer and the evidence it produces.
Adults returning to study — often with families, jobs and very little uninterrupted time — who meet a single hard barrier such as academic English on the way into the workforce. Weekly hours in a lecture hall are not realistic for them. A personal digital tutor that reads their gaps and their available time is, and the avatar can be adapted to the cultural and gender norms of the campus or community it serves.
Neither buys content. Both buy a measurable change in what a defined population can do, inside a budget and a framework that already exist. In both, the human stays: the engine takes the differentiation and the administrative load, and hands back the mentoring, the judgement calls and the confidence-building.
07 — How a pilot works
Before the engine has a track record in a new domain, asking a partner to take our word for it is not reasonable. So the first engagement is built as a measurement, and our commercial outcome is a function of what it measures.
What is proven, what is architectural, and what is still ahead.
Proven: the outcome record above, in Israel's school system, principally
in mathematics and the sciences; delivery verified in Hebrew and in English.
Architectural: domain-agnosticism, language-agnosticism and white-label
deployment are real properties of the engine — properties, not results. A
business-language content library in any further language is built together with a
local partner.
Ahead: the engine's record with adult and workforce cohorts
begins with its first pilots in those cohorts, and we will publish their numbers
the same way we published the school ones.
