Kubeflow Training and Deployment Poller Flow orchestrates end-to-end ML model lifecycle management using Celery Beat schedulers on OCI. Training Poller monitors JOB_CREATED status every 60 seconds, dispatching check_training_status tasks per eligible job and handling failures (budget exceeded, 90-min timeout) or completion. Upon TRAINING_COMPLETED status, Deployment Poller takes over, registering models in MLflow and probing inference servers for up to 30 minutes before marking DEPLOYMENT_SUCCESS or DEPLOYMENT_FAILED. This two-stage polling architecture decouples training supervision from deployment validation, enabling reliable handoff to automated evaluation pipelines. Fork and customize this diagram on Diagrams.so to adapt polling intervals, timeout thresholds, or add model registry integrations. The 60-second beat interval balances responsiveness with resource efficiency for production ML platforms.