Taking ML models to production for logistics at DFDS

Context

DFDS had an AI/ML department building models for the logistics business, but a model only creates value once it runs in production against live data. Getting models off the data scientists’ machines and into reliable production services is its own engineering discipline, separate from building the model in the first place.

What I did

I set up a software team inside the Developer & Platform Experience department to take machine-learning models into production for logistics, working together with DFDS’s AI/ML department. The AI/ML department built the models; my team built the engineering path to run them in production on the cloud-native platform, on the same self-service and golden paths as the rest of the estate, so models shipped and operated as proper services rather than one-off scripts. This was the API development and data-models team inside the department.

Outcome

ML models for logistics could move from the AI/ML department into production reliably, on the shared platform, which gave DFDS a repeatable way to operationalise machine learning rather than a string of experiments. It is the same model-to-production discipline I went on to lead at AXON Networks.