Machine Learning Engineer

Say a company wants to route support tickets to the right team automatically. A data scientist might explore past tickets and show that a classifier could work. The ML engineer makes it real: builds the training pipeline, trains and validates the model, packages it behind an API, watches its accuracy on new tickets, and retrains it when the ticket mix changes. That operating discipline is MLOps.

Typical tools include training frameworks, a model registry, model serving infrastructure and monitoring for model drift and training-serving skew. The work is judged on whether the model performs on live data, not only on a held-out test set.

In episode 12 at 19:10, Google DeepMind’s Logan Kilpatrick recalls working as an ML engineer on computer vision, when one domain-specific problem took nine to twelve months: collect data, train end to end, then build production infrastructure. His point is that foundation models have shortened that path. In episode 37 at 17:21, Hamel Husain argues that people building on models arrive at the ML engineer’s skill set anyway, through error analysis and debugging, even when they never train a model. Both episodes were sponsored by Galileo: Conor points listeners to it in episode 12 at 18:07, and Galileo’s CTO co-hosts episode 37 (0:35).

An AI engineer usually starts from someone else’s model and builds a product around it. An ML engineer more often owns the model itself. Many teams need both, and titles overlap.