Lynx uses multiple agents to improve natural language to sql accuracy
LYNX: Analytical Program Synthesis via Grounding and Closed-Loop Refinement
Artificial IntelligenceDatabases
Summary
It can be hard for computer programs to turn plain English questions into the right database queries, especially when data is complex. The authors present LYNX, a system that breaks down this task into smaller steps with multiple cooperative agents. These agents share the jobs of planning, checking, and fixing mistakes to make translations more reliable and correct. Tests show LYNX works better and uses fewer computing resources than previous methods, and it can handle different styles of underlying language models. It also performs well on finance and complex database tasks, showing it could be useful in real-world data analysis.
What this means in practice
- •For business intelligence teams: Generate more accurate and reliable database queries from natural language descriptions to speed up data analysis.
- •For financial analysts: Execute complex multi-table SQL queries from natural language to support quantitative finance analytics.
- •For software developers: Integrate LYNX to improve natural language interfaces that convert user intent into executable programs with high correctness.$Commercial implications: Makes commercial natural language to SQL products more robust by using multi-agent verification and correction, reducing errors and resource use.
Authors
Ruichen Xia, Xiao Han, Dingyu Yang, Luwei Yang, Sai Wu, Dongxiang Zhang, Binhui Yao, Gang Chen
Abstract
LLM-powered data analytics aims to translate natural language instructions into executable analytical programs over underlying data. However, existing approaches, whether treating LLMs as intent-to-code translators or as standalone analytical engines, often fail to reliably align domain semantics with complex schemas and heterogeneous metadata, resulting in brittle execution and semantically inconsistent analytical outputs in real-world workloads. We present LYNX, a multi-agent framework that formulates LLM-driven analytics as a schema- and knowledge-constrained analysis problem rather than a single-pass generation task. LYNX organizes analytical synthesis into a closed-loop workflow of planning, generation, verification, and refinement through three role-specialized agents that collaboratively enforce schema grounding, semantic consistency, and execution correctness. To address domain ambiguity and schema misalignment, LYNX integrates Semantic-Structural Distillation and Stage-Aware Allocation (SSD-SAA). A delta-guided iterative revision continuously diagnoses and repairs semantic and execution-level failures, substantially improving synthesis robustness. We evaluate LYNX on two NL-to-SQL benchmarks, BIRD-SQL Mini-Dev and the more challenging Spider 2.0-Lite. Experimental results demonstrate strong performance and efficiency, achieving 10.66% to 59.54% execution accuracy across lightweight and commercial models, while reducing token consumption by 70.53% compared with the prior most efficient method. Our framework displays high generalization over different backbones, from open-source, lightweight models to the latest commercial models. Additional case studies on quantitative finance analytics and complex multi-relational SQL workloads further demonstrate the framework's robustness, scalability, and practical applicability in real-world analytical scenarios.