Automatic calibration of building performance simulation models is essential for creating digital twins and training adaptive AI control agents. This paper introduces an automatic multi-phase calibration methodology for a simplified simulation model of a complex multi-purpose building. A multi-objective optimization framework, implemented using the IDA-Python API and based on the NSGA-II algorithm, is utilized to minimize the discrepancies between measured data and simulated outputs. The methodology contains 1) a monthly calibration stage (stage I) and 2) a continuous hourly calibration stage (stage II). First, a monthly calibration is conducted to establish baseline monthly accuracy. Then, an hourly calibration is performed on a weekly basis to refine the model further, ensuring its readiness to be used for training AI control agents and making control decisions. For each week, if the inherited schedules violate acceptable error thresholds between the simulation results and measurements (CVRMS >30% for all objectives), AHU’s fan and internal heat gain schedules are automatically modified to restore hourly accuracy (recalibration). The hourly CVRMSE values for the calibrated model in the whole simulation period were 18.8% for AHU heating and 14.1% for space heating energy consumption. Hence, the model serves as a reliable environment for assessing robust control and management strategies for the real building. This automatic adaptive calibration method can effectively address building dynamics, internal heat gain variations, and sudden weather events such as heat waves or cold waves.