A retail company wants to use Amazon SageMaker for a machine learning project involving both real-time recommendations and batch processing. The company needs to store features with low-latency access for real-time predictions and less frequently accessed features for batch inference. Which Amazon SageMaker capability should they use?