From 700 Trillion Signals to Action: Building Production AI Agents for Real-Time Media Planning

Programmatic advertising teams drown in data signals yet struggle to synthesize coherent media strategies. This session reveals how MiQ engineered a production-grade AI agent using Databricks Unity Catalog, Genie workspaces, and MCP integration to transform fragmented intelligence into prescriptive media plans. We'll demonstrate our ReAct-pattern agent that autonomously orchestrates queries solving the "Inference Gap" between data retrieval and strategic action. Key technical deep-dives include:
 (1) Building reproducible yet adaptive LLM workflows in LangGraph with Databricks SDK,
 (2) Context persistence optimization
 (3) Multi-layered evaluation
Attendees will learn battle-tested patterns for orchestrating Databricks workspaces as agent tools and reasoning engine, achieving enterprise-grade reliability from non-deterministic LLMs, and reducing media planning cycles from days to minutes while standardizing expertise across teams.