Intelligent Support for Collaborative Learning in Advanced Computer Science Courses — From Worked Examples to Data-Driven Insights

In this dissertation, we challenge the preeminence of problem-solving practice in the pedagogy of advanced computer science courses by comparing it to worked examples. In doing so, we not only extend the theory of example-based learning to this novel context, but make significant technological contributions to the infrastructure for conducting AI-facilitated educational experiments at scale. To conduct these experiments, we integrate the Bazaar AI Conversational Agent framework into two courses on the Sail() platform, thus enabling the rapid deployment of AI-facilitated interventions and continuous data collection to support their experimental evaluation.

The findings reveal, in line with theory, that maximizing time spent with worked examples improves conceptual learning. Some problem-solving practice remains necessary for implementation proficiency, especially when learners lack prior procedural knowledge, a novel finding for example-based learning in this context. In addition to the theoretical contributions, these results support a shift in the pedagogy of advanced computer science toward worked example-based reflection, with the optimal design varying based on learners’ prior knowledge.

To fulfill the promise of more effective team-programming project designs, we build on these results to evaluate two strategies sensitive to learners’ prior procedural knowledge levels. The first employs reinforcement learning-based parameter optimization to determine the optimal balance of time spent on problem-solving practice and worked examples. The second employs clever instructional design to create new tasks that simultaneously promote reflection and problem-solving practice. Both designs outperform existing team programming projects and approaches that are not sensitive to learners’ prior procedural knowledge levels.

Collectively, this dissertation achieves three key outcomes: evidence-backed AI-facilitated improvements to team programming project designs in advanced computer science education; theoretical advancements in example-based learning and adaptive pedagogy; and a challenge to conventional teaching practice through empirically validated alternatives. We establish a scalable technological infrastructure to enable continuous experimentation with AI-driven learning innovations, while providing actionable guidance for practitioners and researchers aiming to design adaptive, context-sensitive educational interventions. These contributions collectively redefine the pedagogical possibilities for advanced technical domains, demonstrating how AI integration and iterative experimentation can systematically bridge theory and practice in modern computing education.