AI agents fail in production not because models are weak, but because the data stack is. Fragmented silos, inconsistent definitions, and logic buried in tribal knowledge leave agents overwhelmed with data yet starved for context.

Apache Iceberg fixes the foundation. ACID transactions, time travel, and schema evolution turn storage into a live, versioned context layer that agents can reason over reliably. dbt fixes the meaning. Semantic models and the dbt MCP server give agents a governed interface to your data – translating natural language into structured queries grounded in business logic.

Together, they power Structured RAG: retrieval that understands schema, respects governance, and returns interpretable results.

This session includes a live demo of a fully open-source Structured RAG stack built on Iceberg and dbt, featuring semantic query translation, hybrid retrieval, and governed agent reasoning via the dbt MCP. Expect architecture diagrams, real code, and practical guidance.