Recommender systems are essential for information filtering and retrieval, but their performance is impacted by issues like cold-start and data sparsity. Recent studies have improved recommender system performance using methods like matrix factorization, and various AI-enabled machine and deep learning algorithms. These methods aim to overcome limitations in traditional models, which focus on content descriptions and neglect user-item interaction. In the proposed methodology, an AI-based model that is optimized for Jordan Neural Network is designed as a recommendation system for users to suggest products. Product reviews are pre-processed using techniques like stop word removal, tokenization, stemming, and lemmatization, then fetched into a BERT transform model and reduced dimensionality using LDA. The Jordan Neural Network is optimized to categorize ratings and recommend them to users, with the weight factor optimally selected using the Lyrebird optimization algorithm. According to simulated research, the proposed AI-based product recommendation approach achieves 94.23% accuracy, 5.76% error, and 89.52% precision. The developed technique outperforms existing methods in performance, resulting in a better prediction of product recommendation models.