AIEngiMentor: Scalable Deep Learning System for Hands-on Engineering Education

In this research, we proposed AIEngiMentor, a scalable system that assists students by automatically detecting and diagnosing errors in robotics and engineering assembly. It integrates deep learning models for multi-stage video segmentation and recognition, leveraging the Segment Anything Model 2 (SAM2) to handle both device- and component-level complexities. AIEngiMentor incorporates a Video-based Adaptive Integrity Algorithm (VAIA), which dynamically adjusts to step-by-step requirements from the Learning Management System (LMS). This architecture delivers real-time, precise feedback, even when processing low-quality or moving images with complex backgrounds. Unlike costly industry solutions, AIEngiMentor provides an affordable, practical alternative, ensuring the accuracy and safety of circuit and device connections by validating each step against LMS-defined criteria. Preliminary results show AIEngiMentor outperforms manual inspection in accuracy and enhances student learning by enabling focus on complex problem-solving tasks. The system's adaptability and precision also indicate potential for broader use in industrial automation and healthcare, where real-time error detection is critical.