Exploring the Impact of Different Assistance Approaches on Students’ Performance in Engineering Lab Courses

The rise of large language models (LLMs) offers new forms of academic support for STEM students engaged in self-directed study. This study evaluates the impacts of multiple assistance approaches on laboratory course performance, focusing on engineering students in electronics-related disciplines. A cohort of 218 students underwent a redesigned lab course, and their outcomes were compared to those of 177 students from earlier years who did not receive such support. Specifically, we implemented five types of support approaches for students completing laboratory coursework: (1) Teaching Assistant (TA) only, (2) Generic-LLM-only model, (3) Expert-tuned-LLM-only model, (4) TA + Generic LLM model, and (5) TA + Expert-tuned LLM model. Our key findings are as follows: I. Compared to the historical baseline with no support, students assisted by the generic-LLM-only model did not show a significant improvement in performance. II. Teaching assistant involvement was associated with marked improvements in student outcomes, and performance across all TA-involved approaches showed little variation. III. The expert-tuned LLM was more effective than the generic LLM in improving student outcomes. IV. The combined TA + LLM configurations enhanced learning efficiency overall, although they required greater time investment in the early stages of the course. These results highlight the promising role of LLM technologies in the future of engineering education, while also underscoring the continued importance of domain-specific expertise in delivering effective learning support.