System Design: Data Infrastructure for AI/ML Systems

Design the data path that connects source events, reproducible training datasets, and online features. This lesson defines the workload and checks whether its storage, processing deadline, and serving estimates use compatible units.

A model can be trained on the wrong historical data or served stale features even when its inference endpoint is healthy. The data platform must record where values came from, which transformation produced them, and when they were available.

We will design a shared platform for ingestion, transformation, feature storage, and access by several models. Batch throughput and online latency need separate budgets.

The need for data infrastructure for AI/ML