This paper presents an AI-driven framework designed to enhance user engagement and optimize catalog management in digital libraries. The framework integrates Variational Autoencoder (VAE)-based personalized recommendations with Adam optimizer and Lookahead mechanism for catalog optimization. The VAE model effectively learns latent representations of user-item interactions, providing personalized content recommendations. For catalog optimization, the Adam optimizer with Lookahead stabilizes convergence and refines inventory selection, leading to more efficient resource allocation and reduced costs. Experimental results from a large-scale dataset demonstrate that the proposed approach outperforms traditional methods, achieving significant improvements in recommendation accuracy and user engagement. It reduces the number of low-demand items while enhancing overall catalog efficiency. The proposed framework provides a scalable and adaptable solution for digital libraries, ensuring both user satisfaction and effective resource management. Future work will explore hybrid models incorporating Natural Language Processing (NLP) to improve content understanding and further enhance recommendation quality.