This chapter proposes a guideline-concordant two-stage intelligent system designed to support both the diagnosis and pharmacological management of diabetes. In the first stage, diabetes severity is classified using two distinct datasets—the Pima Indian Diabetes dataset and a local dataset from Iraq—after feature selection and data balancing, applying multiple machine learning and deep learning algorithms with cross-validation. In the second stage, a hybrid drug recommendation module integrates rule-based logic derived from the national diabetes treatment guideline with supervised learning models, incorporating patient-specific variables such as age, body mass index (BMI), glomerular filtration rate (GFR), and comorbidities. The approach achieved high predictive performance in both severity classification and pharmacotherapy recommendation, with random forest and long short-term memory models performing best in their respective categories. By combining clinically interpretable rules with data-driven models and validating on heterogeneous populations, the proposed framework addresses limitations of prior single-task or purely algorithmic approaches and demonstrates potential for real-world clinical decision support. Future work will focus on expanding the dataset, external validation, and deployment in clinical settings to ensure generalizability and safety.