A/B Agent: A Self-Evolving Agent for Strategy Iteration in Industrial A/B Testing

A/B Agent automates industrial recommendation tuning by organizing historical experiment data into a hierarchical structure of business scenarios, stages, and objectives. This architecture enables multi-step strategy refinement through sequential feedback loops, addressing the limitations of flat RAG-based retrieval in cross-scenario knowledge transfer.