From "What Happened?" to "What Will Happen?"

A multi-agent architecture combining Databricks Genie, TabPFN (a foundation model for tabular data), and Agent Bricks enables business users to ask predictive questions in natural language without any ML pipeline setup. Genie translates natural language into SQL to extract labeled training data from the Lakehouse, TabPFN generates predictions in a single forward pass, and an orchestrator supervisor delivers actionable recommendations. The system eliminates traditional data science bottlenecks like feature engineering, model selection, and hyperparameter tuning. An evaluation harness built on MLflow is included to validate prediction reliability across different question types. A domain-agnostic solution accelerator is available on GitHub covering sales analytics, healthcare risk scoring, fraud detection, and more.