From Analytics to AI: Evolving a Data Platform to Power Machine Learning at Scale

Jean-Matthieu Saponaro, Engineering Director at Datadog, has spent 10 years there, from data analyst to leading the company's Analytics Data Platform.

As AI becomes a core pillar of product development at Datadog, the Analytics Data Platform team has had to evolve to support it. Originally built for traditional analytics, business intelligence, and ETL pipelines, the platform now also powers the work of AI engineers, a newer first-class citizen among the data practitioners it supports, building and deploying machine learning models across more than 50 parts of the Datadog product suite.

In this talk, he shares how the team adapted the platform to meet the needs of AI engineering at scale, without compromising on self-service, governance, or reliability. Along the way, they introduced support for new storage formats like Iceberg and distributed processing engines like Ray, moving away from Scala-based Spark toward Python-first tooling that fits how AI engineers actually work.

From self-serve data intake to flexible exploration workflows with robust metadata and observability, the platform experience was reimagined for iterative model development across trillions of data points in a complex, multi-source environment. He walks through how AI engineers at Datadog discover, explore, trust, and use a wide range of data — from observability telemetry to cloud and business datasets — through a unified catalog spanning the Iceberg-based lakehouse, internal time series systems, and the storage powering core products like Logs.

Whether modernizing an analytics platform or building new capabilities for AI teams, this session offers concrete lessons on building scalable, reliable, self-serve infrastructure for data and AI.