Designing educational chatbots in resource-constrained environments: A design science approach

Abstract

Generative AI is expanding the role of educational chatbots in higher education, but many institutions still face financial, computational, and organizational constraints that limit the deployment of advanced systems. To address this challenge, this study uses a design science approach to develop and validate a constrained-environment educational chatbot design and operations framework for introductory human resource development (HRD) courses. The framework combines a resource-efficient small language model (SLM) with fine-tuning, retrieval-augmented generation, and prompt engineering, and is theoretically grounded in the process logics of Factory Physics, Lean, Agile, and Six Sigma to stabilize the learning flow, manage variation, and improve instructional responsiveness. The framework is instantiated through EduHRDBot and evaluated in an authentic university classroom setting for performance, efficiency, and learner experience. The findings show how design science can generate transferable design knowledge for building cost-efficient, pedagogically aligned educational chatbots in resource-constrained environments. Beyond a single-course implementation, the proposed framework offers a modular basis for broader human–AI co-learning ecosystems in higher education.