Test data for a streaming pipeline, from a settings file rather than a program: a plain YAML file describes a simulation, and one command runs it and sends its events to Kafka, PostgreSQL, Parquet or Iceberg.

Generating test data that changes while it runs: SimPy simulations written with a declarative API that read settings from and write events to PostgreSQL and Redis.

One simulation that writes Parquet for model training and streams Kafka events for live inference, so training and serving use the same simulation code.

Keeping a simulation in step with live data: change a running SimPy model's parameters from Kafka and stream its events out, so it acts as a digital twin.

Why a simulation is not a digital twin: the layers that separate an offline model, an operational twin and an event-driven hybrid of the two.

An Apache Flink and Kotlin streaming architecture where online machine learning detects concept drift from machinery wear, controlled by a shadow mode router.