MOSAIC

Introductory programming classes span students who have coded for years and students who have never seen a loop, and motivation varies as much as knowledge. A single static worksheet bores some and loses others: without personalization, 25–30% of low-knowledge students never finished.

MOSAIC builds on FACET, developed by Gonnermann-Müller and colleagues at the Weizenbaum Institute, which uses LLMs to generate differentiated worksheets from an instructor’s own materials. MOSAIC extends it into a richer learner model that combines knowledge, motivation, and skills, and varies scaffolding, explicitness, and tone for each student while the instructor keeps control of the content.

In a first study with 409 CS1 students learning regular expressions, personalized worksheets lifted completion above 99% in every profile. The low-knowledge, low-motivation group gained 18.2% in correctness, a statistically significant result, while perceived challenge stayed the same: personalization acted as a retention scaffold, not a shortcut. The project is now extending its learner model to neurodivergent learners.