Reliable Data Science Analysis with Large Language Models via Multi-Agent Tools Orchestration

WhileLargeLanguageModels(LLMs)showpromiseforautomating the labor-intensive process of data science analysis, their practical application is undermined by the generation of erroneous and unreliable code. We argue that this stems from treating LLMs as open-ended code generators—a task akin to answering an essay question. Weproposeafundamentalparadigmshift: ourframework reframes the task as one of structured tool selection and parameterization, effectively turning the “essay question” into a sequence of “multiple-choice and fill-in-the-blanks” problems. This shift dramatically reduces the potential for error. Our contributions are twofold. First, a multi-agent framework orchestrates the workflow, breaking down complex tasks into verifiable steps. Second, we construct an auto-generated tool library that supports a novel mechanism, empowering LLMs to select tools based on full source code rather than descriptions.

Authors

• Mingyin Zou Strategic Assessments and Consultation Institute, Academy of Military Science
• Guangrong You Strategic Assessments and Consultation Institute, Academy of Military Science
• Xiaomin Zhu Strategic Assessments and Consultation Institute, Academy of Military Science
• Yanqing Ye Strategic Assessments and Consultation Institute, Academy of Military Science
• Ji Wang Laboratory for Big Data and Decision, National University of Defense Technology

Keywords

• LLM
• Code Generation
• Multi-Agent
• Data Science