When a critical supply chain breaks down, qualifying a new supplier often takes weeks of manual audits and verifying – often while small manufacturers with the right machinery sit idle. 

A new three-year National Science Foundation (NSF) grant aims to solve this disconnect by using generative AI to assemble resilient supply chains in minutes instead of months. Department of Computer Science Assistant Professor Xiaorui Liu is leading NC State’s portion of the project, “Generative AI for Autonomous Composition of Resilient and Explainable Service-Oriented Supply Chains.” Farhad Ameri, an associate professor at Arizona State University, serves as the project’s principal investigator.

Traditional industry software optimizes established networks, but it cannot identify alternative suppliers outside the existing system. Liu’s research takes a different approach by composing entirely new supply chains on the fly. 

To achieve this, the project builds on three integrated technical pillars. First, machine-readable data models structure manufacturing capabilities so software can read what equipment can actually produce. Second, generative AI models are grounded directly to a specialized knowledge graph to guarantee that suggested suppliers are real and operational. Finally, the framework measures resilience upfront during the initial composition phase, preventing failure before a supply chain is ever set in motion. 

The concept grew out of research by Liu and Yunqing “Connie” Li, a 2024 NC State CS Ph.D. graduate now at Lenovo, along with former NC State faculty member Binil Starly. In their study published in the Journal of Manufacturing Systems, the team used graph neural networks to predict a manufacturer’s unstated capabilities. 

“Capability didn’t have to be declared to be known, it could be discovered,” Liu said. “If capability can be discovered at the service level, can a whole service-oriented supply chain be assembled based on those inferred decisions? That’s what this project addresses.” 

Developing the platform requires combining generative AI, knowledge representation and physical manufacturing expertise. Liu notes that NC State’s Applied AI in Engineering and Computer Science initiative provides the ideal environment for this cross-disciplinary work. 

“Most open AI problems worth solving sit at the boundary between a strong technical foundation and a domain where the answer matters.”

In the spirit of fostering campuswide collaboration, the Department of Computer Science hosts a biweekly NC State CS AI Seminar Series, connecting computer science researchers with colleagues across materials science, chemistry, energy, life sciences and agriculture.