Self-Serve BI, and Dashboards You Don't Have to Build

A query engine is great for engineers. Everyone else wants a chart. So once the lakehouse was real, the next gap was the part people actually look at: BI. I run Apache Superset over the lakehouse, and I run it the same way I run everything else, as a real platform.

Running Superset like a platform

Superset lives in staging and production on Kubernetes. There are three pieces, each its own repo:

• A forked Helm chart, tracking upstream Apache Superset with our own values and a few template changes.
• A custom image built on upstream Superset with the Dremio driver, SSO, and our dashboard assets baked in, so the container comes up ready to talk to the lakehouse instead of being configured by hand.
• A Helmfile pipeline that pins the chart and image versions and ships both environments through CI.

Once it's up, it connects straight to the lakehouse. People point Superset at a dataset and explore, no ticket, no extract, no copy of the data.

The part I like: dashboards you don't build

Exploration is good, but the bigger win is removing the blank-page problem. Most people don't want to start a dashboard from scratch. They want to see the obvious charts for a dataset and tweak from there.

So I added a one-click generator. For a given dataset it profiles the columns, picks sensible charts (counts, trends over time, top categories, a sample of rows), and assembles a starter Superset dashboard. Then it drops the link to that dashboard into the dataset's page in the catalog, right next to the docs.

That closes a loop I cared about: you find a dataset in the catalog, and a working dashboard is one click away from the same page. Discovery and visualization stop being separate chores. The generation itself rides on the same catalog tooling I write about in the data steward post.

Why it's worth owning

BI tools are easy to stand up and easy to let rot. Treating Superset like a versioned, deployed platform (forked chart, custom image, CI to two environments) means upgrades are routine and the thing people depend on doesn't drift. And the one-click dashboards mean the platform meets people where they are: not everyone writes SQL, but everyone can read a chart.