Full talk · edited transcript

BI Locked Semantic Models are Dead! Open Semantics Just Entered the Chat

David Schwarzenbacher, Maximilian Kohnen, Nehmé Tohmé16 Jun 2026

Speakers: Nehmé Tohmé (Databricks), Maximilian Kohnen (Coca-Cola HBC), David Schwarzenbacher (Databricks)

The biggest blocker in fully adopting Unity Catalog (UC) is not data access; it is the business logic that never makes it there. KPIs, measures and dimensions live inside proprietary semantic models, hand-crafted by analysts over years. In the AI era open semantics are crucial to go the next mile beyond static and manual data analytics, but rebuilding them manually is expensive, fragile and does not scale. The talk shows a scalable, automated and governed solution built with the open-source agentic framework Kasal. Coca-Cola HBC used Kasal on one of their most complex Power BI reports to automatically generate 25+ UC Metric Views encompassing over 200 measures from a single semantic model, powering a fully optimized Genie space on top of it. The same work enabled their IDOR (Insights Discovery On Report) initiative.

The talk has three parts. Nehmé covers the foundation: the agentic framework, LLM calls, agent configuration, guardrails and human in the loop. Max tells the journey at Coca-Cola: the business problem and the pain points. David, who worked closely with the account, walks through the internals of what was built and demos it. The team (all from Switzerland) started the project two years ago and for the past year and a half has focused heavily on migrating business logic into the data layer. The project started before vibe coding; vibe coding and Claude Code then accelerated it a lot.

Continue reading — it's free

Google or a one-time email code. No password.