What are the stages of a production ML pipeline?

Ingest: pull the data from its sources, with a fixed snapshot or time range.
Validate: check schema, nulls, ranges and distributions; stop if the data is bad.
Prepare: compute the features with the same code used for serving.
Train and tune: run with tracked code, data version, parameters and seed.
Evaluate: compare with a baseline and the current model on a fixed holdout, including key slices.
Register: store the model with its metadata and lineage in the registry.
Deploy and monitor: release gradually, watch the service, inputs, predictions and quality, and retrain or roll back when needed.