displayName
ML Model Lifecycle
description
Continuous cross-team workflow managing a machine learning model from initial research
and experimentation through production deployment, monitoring, and eventual retirement.
Data scientists develop and evaluate candidate models in experiment tracking environments,
comparing performance against baselines. The best candidate undergoes code review,
bias and fairness assessment, and reproducibility checks before being handed to ML Ops
for productionisation. ML Ops engineers package the model, configure serving infrastructure,
set up performance and drift monitoring, and deploy via a canary strategy. Model
performance is tracked continuously, with automated retraining pipelines triggered when
drift thresholds are breached. Retirement is planned when the model is superseded or
its use case is deprecated.
workflowKind
data
triggerType
continuous
typicalCadence
per-model
complexity
complex