This paper presents an integrated system, called Personalized Health Assistant (PHA), that utilizes IoT devices and machine learning to monitor diabetes and promote preventive healthcare. For diabetic patients the IoT devices measure glucose levels, heart rate, SpO₂, and physical activity, while non-diabetic patients had their parameters evaluated based on heart rate and stress levels. Models such as Long Short-Term Memory (LSTM) and Multiple Back-Propagation are used to predict continuous variables, while a Convolutional Neural Network is used to analyze dietary images. Additionally, the system integrates a functional chatbot intended to provide efficient health monitoring and decision support for medical professionals. Among the machine learning models explored, the LSTM network achieved highly satisfactory performance in predicting continuous glucose levels. The model reached a coefficient of determination (R2) of 0.9875, with a Root Mean Squared Error of 0.0235, and an accuracy of 94.56% in classifying glucose into clinical categories such as hypoglycemia and hyperglycemia. These results highlight the model’s effectiveness in both regression and classification tasks within a continuous monitoring context.