A customer ticket ran to six lines. What it asked for turned into an architecture spanning Lakehouse Federation, Unity Catalog, AI/BI dashboards, Delta tables and tenant isolation, running inside a production B2B SaaS application.
This session is the field report on that. Not the version where every decision looks deliberate, but what was built, what was measured, what broke, and what got thrown away and rebuilt.
What the session covers
- Lakehouse Federation as the way in, getting live application data into Databricks without building a pipeline first
- Cutting dashboard query overhead, and consolidating several reports into pages
- Tenant isolation for embedded users, which is the part that has to be right before anyone outside the building sees a dashboard
- Moving from federated queries to Delta materialisation, and the 5.4x to 6.7x it bought in dashboard performance
- Refresh schedules designed around when the business is actually working, rather than around a cron habit
- A real refresh flow for embedded dashboards
- Serverless cold starts, caching and idle cost, measured rather than assumed
- Unity Catalog ownership and the deployment pitfalls that come with it
Who it's for
Data engineers, software engineers and architects building customer-facing analytics on Databricks. If you have been asked to put a dashboard inside your own product and the question of who can see which rows has started to feel load-bearing, this is the session.
