Maestro: Making Complex Tasks Simple

Maestro is an AI system designed to simplify complex tasks by separating user instructions from a set of explicit requirements. It operates as a dynamic planning system that determines the optimal sequence of actions, such as using LLMs or other tools, to achieve a goal during inference. The system iteratively improves its output by validating against the specified requirements, employing techniques like Best-of-N generation and generate-and-fix loops. Maestro demonstrates enhanced performance on benchmarks for complex RAG systems and instruction-following tasks, outperforming standard LLM calls by ensuring outputs adhere to defined constraints.