Early identification and proactive management of chronic conditions such as diabetes and kidney complications are essential for improving patient health outcomes. This work introduces a Spark-powered digital twin framework that integrates machine learning and deep learning models to enable accurate prediction and real-time risk assessment. Leveraging Apache Spark’s distributed computing capabilities, the system significantly accelerates model training, reducing diabetes prediction time from 16 min to 8–10 s and kidney complication prediction time using TabNet from 8–9 min to 3–4 s. The framework achieves high prediction accuracy–95% and 91% for diabetes using Random Forest and XGBoost respectively, and 84% for kidney complications using TabNet after tuning and ensembling. Patients are categorized into High, Moderate, or Low risk levels based on predictive scores. To support individualized care, a hybrid food recommendation system combining content-based and collaborative filtering methods is integrated, delivering 5% coverage, 89% mean average precision (MAP), and a 70% hit rate. Deployed as a web application, the system offers timely alerts and personalized guidance, showcasing an efficient, scalable solution for intelligent and preventive healthcare.