Demo: PharmaData-Agent: A Specialized Agent for Pharmaceutical Data Analysis

Zihan Guan, Hanyin Wang, Zhongliang Zhou, Qiaohui Zhou, Peining Tao, Junshui Ma

Keywords: LLM Agent, Pharmaceutical Data Analysis

Abstract: Large language models (LLMs)-based agents are transforming many fields due to their great potential in automating complex tasks. Despite great success, their application in pharmaceutical data analysis remains hindered by coarse-grained workflows and potential data safety risks. General-purpose agents often produce unreliable results and operate with high-risk tools, failing to meet the rigorous demands of data analysis in the pharmaceutical domain. To bridge this gap, we developed PharmaData-Agent, an agent designed specifically for pharmaceutical statisticians. PharmaData-Agent automates the data analysis pipeline through a four-step workflow: data parsing, data understanding, data analysis, and data visualization. Furthermore, by orchestrating sophisticated planning and reflection with specialized subagents, PharmaData-Agent transforms complex tasks into a transparent, multistep process. Empirical evaluations show that PharmaData-Agent achieves perfect tool-calling performance and 95.29% accuracy on DaBench, highlighting its potential to empower statisticians with reliable analytical capability.

Submission Number: 57