How Zalando optimized large-scale inference and streamlined ML operations on Amazon SageMaker

This article discusses how Zalando, a large European fashion e-commerce retailer, optimized large-scale inference and streamlined machine learning operations on Amazon SageMaker to improve their discount steering process for over 1 million products.

Specifically, the article covers:

• Zalando's markdown pricing algorithm, which involves discount-dependent forecasting, determining optimal discounts, providing recommendations, and collecting data
• The motivation for streamlining MLOps, including faster experimentation, scalability, and meeting service-level objectives (SLOs)
• The solution architecture using Amazon SageMaker Processing, AWS Step Functions, Amazon S3, SageMaker Model Registry, and CloudWatch
• Automated production workflows for training, model versioning, and inference using AWS services
• Seamless integration of experiments through SageMaker notebooks and the SageMaker SDK
• Conclusion highlighting the benefits of the implemented architecture, such as improved scalability, streamlined experimentation, and reduced operational overhead