The MLOps Engineer is the management-side operator of models in production. Role 2.1 designs the model and the validation metrics. Role 15.1 owns the release gate. Role 11.1 owns configuration identity. This role makes transition, serving, continuous validation, rollback and early-life maintenance actually run — against the last signed purpose, not against “the pipeline is green”.

Standing duties

• Operate only the configuration Role 11.1 has baselined and Role 15.1 has released. No silent weights, prompts or feature-store swaps.
• Run the continuous-validation metrics Role 2.1 specified. If a metric cannot run, recommend to Role 15.1 (release gate) that the hand-off be held (no veto; ROLES13, t269) — that is an IMP defect, not an OPR workaround.
• Treat drift, performance decay and serving failure as risk and incident signals for Roles 1.4 and OPR02 — on a clock measured in minutes to hours, not the next quarterly pack.
• Keep automated rollback as a working treatment, not a runbook wish. Fail closed when purpose or appetite would otherwise be breached.
• Raise ACT02 when a production change (including a vendor model update) would change purpose in fact. Do not “just deploy”.

Accountable to: AI Governance Lead / Head of AI Architecture (or GM: DevOps / Platform)

Primary domain: Operate (serving, continuous validation runtime, rollback). Implements transition pipelines. Does not Direct or accept residual risk.

Linked EN 18286 clauses: 4.4.2, 5.3, 5.3.2, 6.2.1, 6.2.2, 8.8, 8.9.3, 9.5.3

Format: 3 hours of training, EUR 200.