Incorporating smart technologies into composition-technology teaching can open new avenues for innovation in music education. This study integrates edge computing with college-level instruction in composition technology and proposes a smart digital teaching platform that includes knowledge tracking and personalized practice recommendation. The approach builds a knowledge-tracking model using a self-attention mechanism combined with a hypergraph module, and then formulates exercise recommendation as a POMDP-based decision process to deliver individualized practice resources. Experimental results show the proposed knowledge-tracking model exceeds the accuracy of the DKT baseline by more than 3%, while the recommendation model achieves mean metric values above 0.8, indicating recommendations that are more targeted, novel, and diverse. After deployment, students’ course attitude, learning willingness, and interaction with the platform increased by 25.25%, 30.53%, and 37.68%, respectively. Additionally, over 60% of students reported improvements in teacher–student interaction, recommendation quality, course interest, and learning outcomes. Overall, the platform strengthens interactive teaching, supports personalized learning in music courses, and contributes to enhancing instructional quality in higher education.