Skip to content
Sitao Zhang
Sitao Zhang

Sitao Zhang

Lead Data Scientist, AI
Johnson & Johnson
Bio

Sitao Zhang is a Data Science Lead at Johnson & Johnson, where he has spent nearly eight years driving innovation in the healthcare supply chain. His work focuses on applying cutting-edge data science techniques—including natural language processing, traditional machine learning, deep learning, and mathematical optimization—to solve real-world challenges and advance Johnson & Johnson’s Credo-driven mission. Sitao is passionate about bridging advanced technologies with practical impact in cross functional domains. Outside of work, he enjoys staying active with family through running and CrossFit


MAGIC: Multi-Agent GenAI Inspection Copilot

Objective and Business Problem
The Johnson & Johnson Agentic Inspection Copilot System addresses the critical challenge of clinical trial inspection readiness in a dynamic, highly regulated environment. By automating a wide range of inspection activities, the system enhances efficiency, ensures compliance, and reduces manual effort and bias. This robust, scalable solution streamlines inspection processes, proactively identifies audit risks, and elevates real-time inspection readiness across thousands of sites globally, with an estimated financial benefit of approximately $1.5M projected at the PoC (Proof-of-Concept) stage.

Technical Content
The Copilot System employs a GenAI (Generative Artificial Intelligence) multi-agent workflow orchestrated by LangGraph to streamline inspection readiness. Multi-agents handle key functionalities such as autonomous Self-Retrieval-Augmented Generation (Self-RAG), OpenAI-powered LLM (Large Language Model) responses, and critique-refinement loops, ensuring precision and scalability. By dynamically processing sub-activities, integrating real-time feedback, and generating compliance-ready reports, the system delivers a concise, scalable solution for clinical trial inspection readiness.