The adaptive learning engine that can explain every question it asks.
Ask most vendors how their “AI personalization” picks the next question, and the conversation ends. Here it begins: an ability estimate, a forgetting forecast, a knowledge map and a coach build every session — and this page walks through exactly what each one does to the next ten minutes of practice.
Mechanism one: knowing where the learner is
Everything starts with a computerized adaptive diagnostic. Each question is chosen to be maximally informative given the answers so far, so the ability estimate converges in about two dozen items instead of a hundred-question placement exam. From then on, ordinary practice keeps the estimate live — there is no “assessment season”, just a model that never goes stale.
Mechanism two is the forecast. Each concept a learner has practised carries a per-person retention prediction, decaying between sessions at a rate fitted to that learner’s history. When the forecast approaches the danger line, the concept re-enters their queue. That is the entire trick of spaced repetition — done per concept, per person, at workforce scale.
Mechanism three: the knowledge map
Concepts connect by prerequisite, similarity — and confusion. When two ideas are commonly mixed up, the engine schedules them in the same session, deliberately. Interleaving feels harder and produces dramatically stronger discrimination between the pair.
Mechanism four: the coach
On a miss, the engine serves a Socratic hint aimed at the specific error, not the generic topic. Confidence ratings before each answer feed a calibration score, and systematic overconfidence triggers its own nudges.
The session, assembled
Warm-up reviews, core new material, a stretch item or two, interleaved confusables, cool-down. Each zone has a purpose, and each item in each zone is chosen live from the four mechanisms above.
Inside the engine
Per-concept memory forecasts and the next-best question for each learner — the machinery, made inspectable.
Interface shown as an illustration with representative numbers, not a screenshot — the layout is the product’s.
Watch a session build itself.
Bring a sceptic. We run a live session and narrate every choice the engine makes — question by question, reason by reason.
The evidence this page stands on
Questions buyers ask
Is this the same engine across all products?
Yes — one engine serves practice, diagnostics and verification. What changes by product is the content, the reporting and the audience, not the mechanisms.
What theory is the ability estimation based on?
Item response theory — the same psychometric framework behind serious standardized testing — with the model and its assumptions documented on the science pages.
How does the engine handle brand-new content?
New questions enter with prior difficulty estimates, then calibrate from live answer data. Items that underperform — too easy, ambiguous, miskeyed — get flagged for review automatically.
Can the engine be tuned per company?
Cadence, thresholds and mode mix are configurable per organization; the underlying science is not a setting. You tune how aggressive the schedule is, not whether memory decays.
Why should a buyer care about mechanisms?
Because “AI-powered personalization” without mechanisms is a slogan. If a vendor can’t tell you how a question gets chosen, they can’t tell you why the outcome will repeat.
See it on your own content.
Bring one course. We’ll show you the retention curve your current training leaves behind — and what scheduled review does to it.
- 30 minutes, on your calendar — pick a slot here
- Run on your own content wherever possible, not a canned deck
- You see the dashboards, the learner surface and the evidence exports
- No commitment — and pilot data stays yours either way