The classification of diabetes from structured health records can be used to provide health awareness when model outputs are given with the right limitations. In this paper, we introduce the first Diabetes status assessment framework that integrates a comparative machine learning approach and an interactive application workflow, namely "DiaPredict". The original dataset contained 100,000 records. After removing 3,854 duplicate records, 96,146 unique records remained, comprising 87,664 non-diabetes records (91.18%) and 8,482 diabetes records (8.82%). All eight input attributes (gender, age, hypertension, heart disease, smoking history, body mass index (BMI), glycated hemoglobin (HbA1c), and blood glucose) are processed through a common pipeline and evaluated with Logistic Regression, Random Forest, Extra Trees, and XGBoost. Split of 80 to 20, with state 42 being the stratified portion and random. XGBoost obtains accuracy of 97.07 percent, precision of 95.43 percent, recall of 70.17 percent, F1 score of 80.87 percent and receiver operating characteristic area under the curve (ROC AUC) of 97.56 percent on the held-out test partition. Selected pipeline is coupled with a FastAPI backend and React / TypeScript frontend, SQLite persistence, assessment history, trend visualization, rule based educational insights and PDF reporting. This is a study on diabetes status classification and not future risk prediction as diabetes status was recorded in the target label and not when it occurs in the future. The application thus provides health awareness model probabilities as estimates rather than clinical diagnoses.