Adaptive Curriculum Sequencing Across Activity Types

Research Problem

Can AI automatically adjust learning-path difficulty and content sequencing based on real-time student performance signals, across heterogeneous activity types and subject domains?

Technological Innovation

A plugin-based activity engine supporting 11+ interactive types — flashcards, sorting games, branching scenarios, virtual labs, and external integrations including Desmos, GeoGebra, PhET, H5P and code sandboxes. Each activity type reports granular learning signals: per-question time distribution, retry patterns, wrong-answer clustering, difficulty progression, and engagement decay curves. These signals feed adaptive path recommendations, with automatic difficulty adjustment on the roadmap for every subject.

Why This Is Hard

“Granular” means different things for flashcards versus virtual labs, requiring per-type signal ontologies. The adaptive engine must distinguish between guessing, fatigue and genuine skill boundaries with limited signal. LTI 1.3 integrations return heterogeneous data formats that must be normalized without losing pedagogical meaning, and class-assignment optimization is NP-hard with an exponentially growing constraint space.

Success Criteria

Greater than 10% improvement in completion rates versus static paths, measured through A/B testing of adaptive versus fixed-difficulty groups in pilot schools.