OmniTune: A Universal Framework for Query Refinement via LLMs

Numerous studies have proposed solutions for SQL query refinement, where the goal is to make minimal adjustments to an input query to satisfy a given set of constraints. While effective, these approaches typically address specific query types and constraints, whereas, in practice, users may need to refine a diverse range of queries based on their requirements. To address this, we present OmniTune, a universal framework for query refinement. OmniTune features a Refinement Problem Wizard for defining refinement tasks in natural language and a flexible Refinement Engine, which employs an LLM-based multi-agent architecture to support any query refinement problem. We demonstrate OmniTune across various query refinement scenarios using real-world datasets.