Pedagogical Prompt Engineering Patterns: A Vygotskian Approach to adaptive Scaffolding in Computing Education with LLMs

Paredes-Cayllahua E.L. · Publicaciones de la Facultad de Educacion y Humanidades del Campus de Melilla (2026) · DOI: 10.30827/publicaciones.v56i1.36846

Resumen

Introduction: Large language models are increasingly used in computing education to support explanation, problem solving, and tutoring. However, many interactions with generative AI still emphasize answer delivery rather than structured cognitive engagement. This study proposes a pedagogical prompt engineering framework that integrates Bloom’s Revised Taxonomy and Vygotskian scaffolding principles to design instructionally mediated AI interactions. Method: Following Design Science Research (DSR), we formalized a dual-layer architecture that aligns Bloom levels with scaffolding strategies. The catalog of prompt patterns was evaluated through 54 executions derived from six cognitive levels, three computing topics, and three models (ChatGPT, Gemini, and DeepSeek). The execution code did not configure temperature, top_p, penalties, or seed values; outputs were generated with API-default settings and analyzed with descriptive alignment, similarity, and pedagogical-support metrics. Results: The descriptive results show variations across models and cognitive levels. Overall, ChatGPT presented higher values of alignment and scaffolding support, followed by DeepSeek and Gemini. These differences are interpreted as exploratory trends within the scope of the study. Conclusions: The study provides preliminary evidence for designing prompts as instructional artifacts aligned with cognitive objectives and pedagogical mediation. These results should be interpreted with caution and further examined in future studies using additional repetitions, inter-rater review, and real learning data.

Instituciones: Universidad Nacional de San Agustín de Arequipa