A machine learning engineer is tuning an XGBoost model on SageMaker with a large search space of 7 continuous and categorical hyperparameters. Each training job takes about 45 minutes and the team has a limited compute budget, so they can afford only about 30 total training jobs. They want the automatic model tuning job to intelligently use results from completed jobs to guide the selection of hyperparameters for subsequent jobs, maximizing the chance of finding a strong configuration within the budget. Which tuning strategy should they configure?