This research suggests a novel system structure and optimization plan utilizing deep learning to enhance the personalized recommendation system for engineering courses in higher education, with the goal of bettering personalized learning in the education sector. This research creates and executes a mixed model that blends multi-layer perceptron (MLP) and convolutional neural network (CNN) to enhance the adaptability and expandability of the recommendation system via a modular system architecture. By optimizing data processing and recommendation strategies, we implement a hybrid approach combining collaborative filtering and content filtering to greatly enhance the precision and variety of recommendations. Experimental results show that the hybrid method outperforms the traditional method in multiple performance indicators and shows excellent application potential in higher education engineering course recommendation. This system effectively helps students improve their learning efficiency and enhance their learning experience, while providing teachers with data support to develop more effective teaching plans. The research not only enriches the theoretical basis of personalized recommendation systems, but also provides practical guidance for educational technology companies to develop intelligent learning platforms and provides strong support for the innovative development of education systems.