Arul Bharathi

McAfee protects millions of users through products like Scam Detector and Social Privacy Manager. Iteration on adaptive pricing models was bottlenecked by slow, manual feature engineering. To unlock value from massive user signals, we needed to move from ideation to production fast. In this talk, we show how we built an autonomous feature engineering agent on Agent Bricks that cuts engineering cycles by 60% — replacing manual SQL with an agent that reasons, iterates, and converges. This is the next step in a journey where AI-assisted feature store and personalization have contributed to $54M+ in incremental revenue over three years. The agentic approach is how we make that capability reproducible, faster, and accessible without a team of ML specialists. Attendees will see a live blueprint of our agentic pipeline: - Analyze & ideate: Agent loads domain context and playbook from Unity Catalog, analyzes baseline and proposes features via LLM reasoning - Generate & evaluate: Agent writes code, executes in sandbox, evaluates against baseline and logs every iteration to MLflow. - Converge & deploy: Best features publish directly to Unity Catalog feature store. Walk away with a…