data analytics and engineering focused on AI

I have been looking for a deeper explanation of semantic layers, and recently found Cornelius van Heerden’s research on the Semantic Gold Layer and ELTM: Extract, Load, Transform, Model. The idea of making modeling an explicit step in the pipeline connects with how I work: defining business logic in the model so that people and AI tools can use consistent meanings.

The paper also got me thinking about where that semantic layer should live. I have been building it in the modeling layer, while metadata in the data warehouse seems to offer another approach. I am still learning how those designs compare and where each one fits.

I do not see dashboards disappearing. They give people a reliable place to land, check current figures, and work with complex calculations. But a dashboard can also be the start of a conversation: when an aggregation is unclear, people should be able to ask why and trace the answer back to the underlying data.

In my experience, around 90% of user questions rely on a relatively small set of dimensions and metrics. That makes me curious about which questions truly benefit from an AI agent and which are already answered more simply by a well-designed dashboard. For just-in-time reporting, an agent also needs to explain how an answer relates to the data at its lowest useful grain.

I am continuing to read about these ideas and compare them with what I see in practice.