Quality Checks With Bauplan and Claude Code

Quality checks are how you know the data flowing through a pipeline is what you actually expect, before anything materializes to production. In this demo we add them to an e-commerce pipeline using Bauplan and Claude Code.

We start in the IDE with Claude Code, a CLAUDE.md pulled from Bauplan's GitHub repo, and a data exploration file from an earlier session that the e-commerce pipeline was built from. The Bauplan skill is installed as a plugin from the Anthropic marketplace, so Claude knows how to scaffold checks correctly.From there, we hand Claude a query from the data exploration session and start a quality check session.

Claude uses the quality check skill to build an expectations file, using Bauplan decorators and the SDK to validate the pipeline's inputs and outputs.On the first dry run, run locally before any materialization to the lakehouse, Claude catches a real problem: the event type values in the source data do not match the expected enum. It queries the source table, finds the actual values (view, cart, purchase), updates the expectations file to match and runs again.

The result is an expectations file with 10 checks across staging sessions, staging users, and session metrics, covering enums, referential integrity, business rules, and row-count monitoring. The final dry run validates the full DAG with no blocking failures, ready to commit or run in full on the dev branch.