Vertex AI Pipelines — Vertex AI MLOps

Serverless Pipeline Orchestration with Kubeflow (KFP)

Training models inside interactive Jupyter notebooks does not satisfy enterprise repeatability, compliance, or CI/CD mandates. Vertex AI Pipelines executes containerized, reproducible machine learning workflows with zero infrastructure management.

Pipelines are authored using the open-source Kubeflow Pipelines (KFP v2) SDK. Every component execution automatically registers input artifacts, output models, and evaluation metrics into Vertex ML Metadata, establishing an immutable audit trail of model lineage.

from kfp import dsl
from google.cloud import aiplatform

@dsl.component(base_image="python:3.11")
def evaluate_model(test_data_uri: str, accuracy_threshold: float) -> bool:
    # Model evaluation logic
    current_acc = 0.942
    return current_acc >= accuracy_threshold

@dsl.pipeline(name="enterprise-fraud-detection-pipeline")
def fraud_detection_pipeline(threshold: float = 0.90):
    eval_task = evaluate_model(
        test_data_uri="gs://ml-dw/test_data.parquet",
        accuracy_threshold=threshold
    )

# Submit serverless execution to Vertex AI Pipelines
job = aiplatform.PipelineJob(
    display_name="fraud-pipeline-run-001",
    template_path="fraud_pipeline.yaml",
    pipeline_root="gs://ml-dw/pipeline_artifacts",
    enable_caching=True
)
job.submit()