You're working with advertising data to support both AI model development (which requires near-real-time data cleansing) and historical analytics. One of your main goals is to detect longtail patterns and outlier data points. What is the most effective architecture to achieve this?