MLOps
What is MLOps?
MLOps is the operational discipline of keeping a machine learning model working in production - deployment, monitoring, retraining, and versioning - as opposed to building it once.
Why it is a separate discipline
Ordinary software fails loudly. A model fails silently: it keeps returning confident numbers that have quietly stopped being right. Nothing throws an exception when your training data goes stale. That single property is why ML systems need monitoring built for them rather than the monitoring you already have.
The minimum viable version
Versioned model artefacts so you can roll back. A feature pipeline shared between training and serving so the two cannot diverge. Input and output distribution monitoring. A documented retraining procedure someone other than the author can run. A shadow-mode path for new versions. Everything beyond this is optimisation.
What most small teams over-build
Feature stores, experiment-tracking platforms, and orchestration frameworks solve problems you get at a scale most businesses never reach. A single scheduled job, a versioned artefact in object storage, and a monitoring dashboard covers a production model serving one business unit. Adopt heavier infrastructure when a real constraint demands it.
The ownership question
The critical thing to settle at handover is who runs the retraining and who watches the monitoring. A model with no named owner is a model that degrades unnoticed for a year. This is an organisational answer, not a tooling one, and it is worth agreeing before the build starts.
We build automation systems for small teams. Free audit call to map your specific workflows - no pitch, just a plan.