Prompt Engineering for Personalized AI Teaching Assistants

Summary

Large language model teaching assistants can provide scalable educational support, but their responses are often insufficiently personalized. This study introduces a prompt-engineering framework for general-purpose LLM- and retrieval-augmented generation-based assistants, including systems such as Jill Watson, across disciplines and courses. It models learners along six dimensions: self-assessment, abstraction preference, verbosity preference, perceptual orientation, information-processing style, and level of understanding, producing 96 possible learner profiles. The framework also applies Bloom’s Taxonomy to estimate the cognitive complexity of each student query. These learner and query attributes are encoded in structured prompts, allowing the assistant to adapt without retraining the underlying model. Experiments using natural-language-processing metrics and a human study with five participants found perceived differences in response style and structure across personalization conditions. Statistical analyses also identified learner attributes linked to measurable response changes. The authors describe the findings as preliminary evidence that prompt-based personalization can support adaptive behavior in LLM-powered educational agents.