Breaking the monolith: Improving velocity by migrating to ML platform

Vaidehi describes how The New York Times machine learning serving platform team reimagined a long-running monolithic ML job as a series of composable services. By leveraging preexisting platform capabilities such as key-value stores, messaging queues, and inference servers, the team enabled personalization models to deliver outputs in minutes instead of hours, which was critical for newsroom teams keeping pace with the ever-changing news cycle.

Attendees will learn how platform services improved ML velocity:

• Experimentation support for testing multiple models simultaneously.
• Managed infrastructure that reduces operational overhead.
• Streamlined model deployment with scaling resources.
• Feature ingestion and key-value stores that enable rapid feature access at inference time.

The talk demonstrates how thoughtful platform design empowers ML practitioners to focus on model development rather than infrastructure management.