Prompt-Engineering Framework Enables Scalable Micro-Level Personalization for an LLM AI Teaching Assistant

Forum topic · 小凯 · 2026-09-06

Summary

A study by Saptarshi Basu, Sandeep Kakar, and Ashok Goel (arXiv:2509.00005) introduces a prompt-engineering framework that adds hybrid micro-level personalization to general-purpose LLM/RAG-based AI teaching assistants such as Jill Watson, without model retraining. The framework adapts responses along six learner-specific dimensions: self-assessment, abstraction preference, verbosity preference, perceptual orientation, information processing style, and level of understanding, yielding 96 distinct learner profiles. Student queries are also analyzed using Bloom's Taxonomy to estimate cognitive complexity at the interaction level, and learner attributes plus cognitive assessments are encoded into structured prompts to condition the LLM. Evaluation combined NLP metrics with a five-participant human study, showing perceived differences in response style and structure under personalization and statistical associations between learner attributes and measurable response changes. The findings offer preliminary evidence that prompt-based personalization can support adaptive behavior in LLM-driven educational agents, making personalization scalable and flexible across academic disciplines and courses.

Paper Overview

Field: AI/ML Authors: Saptarshi Basu, Sandeep Kakar, Ashok Goel Published: 2026-09-06 arXiv: 2509.00005

Abstract

Artificial intelligence (AI) teaching assistants powered by large language models (LLMs) offer scalable educational support but often provide limited personalization. This study presents a prompt-engineering-based framework for personalizing general-purpose LLM/RAG-based AI teaching assistants such as Jill Watson across academic disciplines and courses.

The framework adapts responses using six learner-specific dimensions:

Self-assessment
Abstraction preference
Verbosity preference
Perceptual orientation
Information processing style
Level of understanding

Together these yield 96 distinct learner profiles.

Student queries are additionally analyzed using Bloom's Taxonomy to estimate cognitive complexity at the interaction level. Learner attributes and cognitive assessments are encoded in structured prompts, conditioning the LLM without requiring model retraining.

Evaluation

The framework was experimentally evaluated using:

1. NLP metrics 2. A human study with five participants

Results showed perceived differences in response style and structure under personalized conditions, and statistical analysis identified learner attributes associated with measurable changes in responses.

Conclusion

These findings provide preliminary evidence that prompt-based personalization can support adaptive behavior in LLM-driven educational agents, enabling scalable, flexible, and real-time micro-level personalization without retraining the underlying model.