Vertex AI Pipelines

Automating the ML Lifecycle

Training a model in a notebook is easy. Deploying and maintaining it in production requires MLOps. Vertex AI Pipelines allows you to orchestrate ML workflows.

A pipeline might look like this:

1. Extract data from BigQuery
2. Preprocess and normalize data
3. Train a custom model
4. Evaluate model accuracy against a baseline
5. If accuracy improves, deploy to a Vertex Endpoint

Pipelines are serverless and defined using the Kubeflow Pipelines (KFP) SDK.