<![CDATA[Rural high schools in low-resource environments face significant barriers to AI-enhanced hardware simulations, including limited network bandwidth, low-specification devices, and a lack of localized offline tools. This exploratory pilot study proposes and evaluates a low-resource open-source AI simulation framework integrating MobileViT for student behavior detection, Grad-CAM heatmaps, and open-source tools such as QEMU, Logisim-evolution, Tinkercad AR, MagicSchool.ai, and Blender. The framework was implemented through WeChat in an 8-week A/B testing intervention involving 25 students from a rural high school in central China. It was optimized for offline compatibility, rural agricultural contextualization, and privacy protection using anonymous IDs. An exploratory statistical analysis was performed to examine the potential mediating pathways. The results showed approximately 30% improvement in learning efficiency, 25% improvement in test accuracy, and 28% improvement in participation rate, with bootstrap-based 95% confidence intervals indicating positive effects, although these should be interpreted cautiously due to the small sample size. Large effect sizes were observed (Cohen’s d > 0.8, p < 0.001); however, their generalizability remains limited in this pilot context. MobileViT showed a preliminary mediating role in increasing participation and reducing cognitive load, consistent with Cognitive Load Theory and Self-Determination Theory. The framework supports UNESCO’s digital equity principles through equitable access, bias minimization, privacy protection, community participation, zero-cost deployment on standard teacher PCs, and a public GitHub repository with CI/CD pipelines. This study offers a practical and replicable preliminary solution for inclusive STEM education in resource-constrained K-12 classrooms globally.]]>
Copyrights © 2027
