From Cognitive Overload to Individualized Learning: AI Tutoring in Electrical Engineering

Abstract

First-year students in electrical engineering often encounter significant challenges in their undergraduate programs, particularly due to heterogeneous prior knowledge in mathematics and physics, as well as difficulties in managing their own learning processes. These issues frequently result in cognitive overload, poor exam performance, and high dropout rates. Traditional teaching methods, especially in large cohorts, struggle to address these challenges effectively, as excessive tutorial time is spent on basic calculation steps rather than fostering deeper, subject-specific understanding. To address this, a domain-specific AI tutor was piloted during the winter semester of 2025/2026 in the first-semester course "Fundamentals of Electrical Engineering". The AI, based on the Google Gemini 2.5 Pro language model, was enhanced with Retrieval-Augmented Generation (RAG) to integrate tailored teaching materials, such as annotated lecture transcripts and detailed solution manuals. Configured as a Socratic tutor, the AI provided individualized, asynchronous support by guiding students through tutorial preparation, verifying their calculation steps, and assisting with step-by-step troubleshooting. This tool was tested with a seminar group of 18 students who got access to the AI tutor via the platform you.com. The pilot project included continuous data collection throughout the semester. Weekly online reflection forms captured students' usage patterns, perceived task difficulty, and feedback on the AI's usefulness or limitations. These data were complemented by two comprehensive surveys conducted at midterm and end of semester. The study presents the technological configuration and didactic applications of the AI tutor, analyzing its impact across different student performance levels. The results indicate high acceptance of the AI tutor, particularly for verifying calculations and resolving short-term learning obstacles. However, the analysis also revealed limitations, especially in handling subject-specific visual and topological contexts. The findings underscore the importance of fostering critical AI literacy among students and highlight the need for future multimodal AI models to better support engineering education. In addition, the use of AI in undergraduate studies serves to build competencies within the context of a potential core curriculum.