STEM education: Understanding secondary students’ epistemic cognition in the design process with the support of a personalized multi-agent system

The engineering design process is a vital component of integrated STEM education, where students’ design practices both shape and are shaped by their epistemic cognition (EC). Nonetheless, research on secondary students’ EC in STEM contexts is limited, and implementing personalized STEM instruction that accommodates diverse students’ interests and needs poses significant challenges. Harnessing a Multi-Agent System (MAS) afforded by Generative Artificial Intelligence (GAI) offers a promising pathway to address this gap. Grounded in the design thinking model of E DIPT (Empathize-Define-Ideate-Prototype-Test), this study developed a P ersonalized MAS , namely EPMAS, to support secondary students’ STEM project-based design process. Twenty students, grouped into dyads, participated in a five-day STEM winter camp. We sought to explore students’ EC within the EPMAS-supported design process and their perceptions of the system. Data were collected through in-depth group interviews, supplemented with student artifacts and interaction log data, and analyzed using theoretical thematic analysis. Findings revealed students’ engagement with epistemic aims, ideals, and reliable processes, reflecting their developed understanding of engineering design. Contrary to common concerns about over-reliance on GAI, students demonstrated high epistemic agency, critically employing EPMAS’s stepwise guidance as part of their epistemic repertoire to pursue creative solutions. These results suggest that EPMAS-supported authentic STEM learning holds promise for broader school adoption.