Sai Yang

Mercedes‑Benz Korea runs its business on a rich set of sales and steering KPIs maintained in Power BI semantic models and a central Master KPI catalog. Over time, this BI‑centric implementation made it hard to reuse KPI logic for new AI experiences and across platforms. To enable “Talk to Your Data” on top of governed KPIs, Mercedes‑Benz Korea partnered with Databricks to move KPI logic into the Databricks Lakehouse and Unity Catalog. In this session, we show how we developed an automated DAX‑to‑SQL transpiler that converts existing DAX measures into Unity Catalog Metric Views, while preserving semantic parity with legacy reports. We’ll walk through the architecture, how we use the Master KPI List as the semantic contract, and how we tuned Databricks Genie with KPI‑aware metadata and instructions. Finally, we’ll share lessons from making AI agents production‑ready—from handling complex DAX patterns to rolling out self‑service “talk to your KPIs” capabilities across markets.

Managing AI at the scale of Mercedes-Benz requires more than just models—it requires a unified AI ecosystem. With thousands of practitioners across global business units, the challenge was providing a single point of entry for AI capabilities while maintaining strict governance across multi-cloud and multi-model environments. In this session, we go under the hood of the Mercedes-Benz. AI Ecosystem. We will detail the technical architecture that integrates Databricks Agent Framework, Model Serving, and AI Gateway to create a governed interoperability for AI agents. We’ll discuss the "how-to" of: - Centralizing discovery: Building an internal marketplace for foundation models and reusable agents - Unified governance: Integrating Mercedes-Benz AI Marketplace into Databricks AI Gateway - Operationalizing agents: Deploying and publishing AI agents for self-service at Mercedes-Benz Attendees will leave with a technical blueprint for building an interoperable enterprise AI agent platform.