Mastery learning with AI: how Alpha Learning in GroutApp works

Alpha Learning is a new module in GroutApp that runs a class on mastery rather than coverage: a student moves on from a concept only when they have shown, through practice, that they understand it. It is available today on institution subscriptions and is coming soon to individual plans.

The idea is not new. Mastery learning has been studied for half a century, and the evidence that students do better when they are not pushed past gaps is consistent. What has always failed is the logistics — one teacher cannot track thirty students on thirty different concepts, write extra practice for each of them and notice who has quietly stopped trying. That bookkeeping is what Alpha Learning takes over.

Start with a map, not a list of chapters

A textbook is linear because paper is linear. Understanding is not. Refraction depends on knowing that light travels in straight lines and that its speed changes between media; a mirage depends on refraction and on how air density changes with temperature.

So an Alpha Learning curriculum is a concept graph. A Guide — our word for the teacher running the class — generates one from a board, grade and subject, optionally pasting in their own syllabus. The result is two to eight units holding 20 to 60 concepts, each one a single skill that can be mastered in one 15–45 minute sitting, linked by prerequisites.

A concept unlocks only when every one of its prerequisites is mastered. The graph can never loop back on itself, and every concept can be edited, reordered or rewired before the curriculum is published.

See the curriculum studio and the rest of Alpha Learning

Every concept gets the same structure

Consistency matters more than cleverness here. A student who knows where the worked example will be spends their attention on the idea, not on navigating the page. Each concept carries:

• Three lessons, always in the same order: an explanation, one worked example solved step by step, and a short summary of what to remember.
• Eight to twelve practice items mixing multiple choice, short answer and numeric questions, spread across difficulty levels, each with an explanation.
• Optionally, a narrated and captioned animation of four to eight scenes: a hook specific to the topic, the core idea, a worked example and a recap.

Wrong answers should tell you something

The weakest part of most generated quizzes is the distractors. If three of the four options are obviously wrong, a correct answer measures nothing and a wrong one tells the teacher nothing.

Alpha Learning asks for multiple-choice options built from real misconceptions — the answer a student gets if they add where they should multiply, or confuse the angle of incidence with the angle to the surface. Short-answer items list every accepted form of the answer, and numeric items carry an explicit tolerance, so a student is not marked wrong for writing 0.50 instead of 0.5.

What “mastered” actually means

Mastery is not a button a student presses after watching a video. Each practice attempt updates a running score that weights recent answers and falls faster after a mistake. A concept is marked mastered only when all three of these hold:

• the score has reached 85%,
• the last three answers were correct in a row,
• and the student has made at least four attempts.

Mastered concepts then come back for spaced review — first after three days, then at growing intervals up to sixty. Fail a review and the concept drops back to learning. When a student sits down, the app chooses what they work on: reviews that are due first, then the unlocked concepts with the lowest score, then syllabus order.

An AI tutor that refuses to do the work

Every session has an AI tutor available, grounded in that concept’s lesson text, the student’s mastery state and their recent wrong answers. Its most important rule is the one students like least: it never reveals the final answer to a practice item.

Instead it gives hints, asks leading questions and, if a student is still stuck, works a similar but not identical example. It addresses the misconception behind the last mistake rather than just correcting it. Each session has a hint budget — twelve by default — so the tutor supports thinking rather than replacing it.

The Guide’s job changes

When the software handles sequencing, practice and marking, the teacher’s time moves to the things software is bad at. The Guide dashboard shows the whole class — average mastery, minutes practised over the last week, streaks, explainers watched — and flags students who are inactive, stuck or below pace.

From there a Guide records check-ins after talking to a student, sets goals, schedules afternoon workshops on life skills and tracks attendance. Guardians see progress through a separate, read-only portal.

Personalisation: a different route, never an easier one

The feature we are proudest of is custom learning. A Guide can commission a version of any topic written for one student, drawing on a learner profile — traits such as “prefers visuals”, “needs very short steps”, “anxious about this subject” or “fast finisher”, plus free-text notes — together with their goal and recent check-ins.

The rule behind it is strict. The student’s version keeps the same syllabus scope, the same number of lessons and practice items and the same rigour. What changes is the examples, the pacing, the wording and which misconceptions are tackled head on. Their animated explainer is rewritten for them too. The rest of the class keeps the shared version, and the Guide can reset the topic to shared at any time.

This is the line we think matters most with AI in schools. Lowering the bar for a struggling student is easy and does them no favours. Finding them a better way to the same bar is what a good teacher does — and until now it has only been possible for the few students a teacher had time for.

Motivation that rewards learning

Points, streaks and badges are paid for the things that build understanding: completing a session, watching an explainer through to the end, finishing a daily plan, attending a workshop and, above all, mastering a concept. The same event never pays twice. A daily plan is built from 25-minute focus blocks with five-minute breaks, and any planned minutes a student does not need come back to them as time back.

Honest limits

Generated lessons are a first draft, not a finished product. The curriculum studio exists because a Guide should read and edit what is published — especially prerequisites, which encode judgement about how a subject is best learned. Mastery tracking also only sees what happens in practice; it does not replace a teacher’s conversation with a student, which is exactly why check-ins feed back into the material.

Availability

Alpha Learning is available now for schools, colleges and institutes on a GroutApp institution subscription, with roles for students, Guides and admins. It is rolling out to individual GroutApp plans soon.

Explore Alpha Learning · Grout for schools and colleges

Frequently asked questions

Is Alpha Learning included in individual GroutApp plans?

Not yet. It is available now on institution subscriptions and is coming soon to individual plans.

What is a Guide in Alpha Learning?

A Guide is the faculty member running a class. Guides review and publish curricula, watch class progress, record check-ins, set goals, run workshops and commission custom versions of topics for individual students.

How does Alpha Learning decide a concept is mastered?

When the student’s practice score reaches 85%, their last three answers are correct and they have made at least four attempts. Mastered concepts return for spaced review between 3 and 60 days later.

Does personalising a topic make it easier?

No. A personal version keeps the same syllabus scope, lesson and practice counts and rigour. Only the examples, pacing, wording and targeted misconceptions change.