Dovetailing Case-Based Reasoning and Large Language Models to Compare Teaching Strategies: A Mnemonic Augmentation Framework using the EEDI Dataset

We present a novel hybrid framework dovetailing case-based reasoning (CBR)and AI teaching personas that facilitates mnemonic augmentation to evaluate different teaching strategies for mathematical misconception reduction. Ourapproach uniquely combines empirical student data from the EEDI dataset (17million student interactions) with AI-driven teaching simulations. Each teaching strategy is implemented as an AI persona with CBR-driven mnemonic aug-mentation that relies on the retrieval and application of historical cases whilemaintaining the ability to generate adaptive responses. Through comprehensive cross-validation, we demonstrate that this hybrid approach achieves a 15% reduction in misconceptions compared to traditional methods, outperforming bothpure CBR (8%) and standard AI approaches (11%). The framework providesa a data-driven foundation for evaluating and selecting teaching strategies while acknowledging the complexity of educational interventions.