Team Topologies: Who Does What

Five recurring roles

• Data engineer — pipelines, schemas, freshness, lineage.
• ML engineer — training code, model packaging, serving glue.
• Data scientist — feature design, modelling, evaluation.
• Platform / SRE — runtime, CI/CD, observability, on-call.
• Product / risk — sign-off, policy, prioritisation.

Anti-patterns to spot

1. Throw-it-over-the-wall — DS hands a notebook to MLE, MLE rewrites everything, knowledge is lost.
2. One platform team for everyone — bottleneck; product teams stop owning quality.
3. No SRE involvement — first incident is the team's first conversation about SLOs.

Healthy pattern (Team Topologies vocabulary): a stream-aligned team owns the model end-to-end, supported by an enabling platform team and an internal platform product.

A working ML org is like a modern hospital: doctors (DS/MLE) treat patients (models), nurses (data engineers) keep them stable, the hospital infrastructure team (SRE) runs power and oxygen, and a clinical-governance committee (product/risk) approves new procedures. Nobody is expected to do all four jobs, but everyone is expected to coordinate.