Knowledge-tracing algorithms and natural-language processing continuously model each student's understanding and adjust the learning path in real time. This runs across content formats and existing LMS integrations.
The challenge
Modelling the complex, evolving nature of student knowledge states. Recommendation that balances learning efficiency against engagement. Coverage across diverse subjects and difficulty levels. And keeping educators firmly in oversight and control of the process, not displaced by it.
The approach
A deep knowledge-tracing model enhanced with attention mechanisms captures fine-grained understanding across interconnected concept hierarchies. A reinforcement-learning agent sequences activities for long-term retention rather than short-term quiz scores, and the system generates natural-language explanations adapted to each student's level.
Teachers saved around five hours a week on individualised planning. That time went back to students.
The results
Across three institutions, the platform now serves over 10,000 active students.
- 28%
- Higher exam scores
- 40%
- More engagement
- 50%
- Lower dropout
Exam scores improved 28% on average, engagement rose 40%, and course dropout fell 50%, with teachers reclaiming roughly five hours a week from individualised lesson planning.
