We explore Google Research's D-Star, a data-science agent that reads heterogeneous data (CSV, JSON, Markdown, and more), extracts structure and context, and turns questions into executable Python code through a plan–implement–verify loop. Learn how a dedicated verifier critiques outputs beyond syntax, how a router can revise or replan to prevent error cascades, and why this self-correcting approach yields state-of-the-art results on benchmarks like DabStep, Kramabench, and DECODE. Practical implications for researchers and policy analysts wrestling with real-world, messy data—and what it could mean for democratizing automated discovery.Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information. Sponsored by Embersilk LLC

Ver todos los episodios