Teaching an AI Agent to Build a Data Warehouse

This talk presents Modlr, a prototype AI agent that automates the design and management of data warehouses, one of the most complex and error-prone tasks in data engineering. Modlr uses Claude Code as its main agent, orchestrated with FastMCP and LangGraph, to reason about structured business context, enforce modeling conventions, and iteratively build production-ready dimensional models and semantic layers. During the talk, I will demonstrate the agent creating a complex warehouse model live—transforming a task that normally takes weeks into under two minutes—and show visualizations via Mermaid lineage and ERD diagrams, immediately deployable to dbt and MetricFlow for querying via SQL or natural language. I will walk the audience through the code, system architecture, technology stack, and validation mechanisms that make this possible.