In any classroom of thirty, some students are bored and some are lost, and the teacher - however skilled - can only be in one place at a time. This project set out to change that, by giving every learner a patient digital tutor and every teacher a capable assistant.
We worked with schools, ministries and non-profits across several continents to build and deploy adaptive learning at real scale - not a pilot that photographs well, but a system used every day by students who need it most.
The challenge
The gap in learning is rarely about effort. It is about pace and support. A student who missed one foundational concept falls further behind with every lesson built on top of it, until the subject feels impossible. Large class sizes and stretched teachers make it almost impossible to catch that moment of divergence in time.
The design brief, therefore, was blunt: find each student's exact gap, help them close it, and do it in a way that works in a crowded city school and in a village with intermittent power alike.
Our approach
- Diagnose before you teach. Every session begins with a short adaptive check that locates the weak prerequisite beneath a wrong answer, not just the wrong answer itself.
- Make the student think. The tutor asks questions and gives hints rather than answers, so learning is earned, not copied.
- Support the teacher. The same engine generates differentiated practice, translates material and flags who needs attention before the lesson starts.
The technology
The tutoring engine combines knowledge-graph models of each curriculum with conversational AI that reasons about a student's answer. If a learner adds two fractions incorrectly, the system distinguishes between a rules error and a conceptual one, then chooses the next exercise accordingly.
To reach low-connectivity settings, the core models run on-device and sync when a connection is available. Assessment is multilingual from the ground up, so a student can read a problem in one language and answer in another while the system tracks understanding across both.
Good tutoring does not answer the question. It asks the next one - and waits.
What it delivers
The results, measured over full academic terms rather than single sessions, point in one consistent direction: the students who start furthest behind make the largest gains, because the system never lets a gap grow silently.
Representative outcomes
- Faster progress for lower-attaining students using adaptive practice.
- More time for teaching as planning and marking are assisted.
- Multilingual access opening curricula to newcomers.
- Offline capability reaching students without reliable internet.
- Earlier intervention through at-risk flagging for teachers.
A global picture
Every deployment taught us something new.
- Estonia treated AI tools as ordinary classroom infrastructure, easing adoption.
- Finland balanced automation with a strong emphasis on teacher judgement and critical thinking.
- India used multilingual assessment to reach learners at a scale one-to-one teaching could never match.
- Across Africa offline-first design delivered tutoring where connectivity cannot be assumed.
- Brazil and Southeast Asia supplemented crowded classrooms with conversational practice.
The human side
A tutoring system is only as good as the trust it earns. Teachers were involved in design from the first week, and every recommendation the engine makes is explainable: why this exercise, why this student, why now. We deliberately refuse features that would let a student skip the struggle, because that is exactly where learning happens.
Privacy was treated as a first-class requirement. Detailed records of how a child learns are among the most sensitive data there is, and they are protected far more strictly than a typical consumer app.
What is next
The next chapter is deeper integration: tutors that carry a student's understanding from year to year, teachers who can see the whole picture at a glance, and assessment that measures thinking rather than the ability to produce an essay a machine could write. The aim is unchanged - a classroom where no learner is left to fail in silence.