Presenter: Jinxin Chen
Organization: Stevens Institute of Technology
STRATEGIC WORKFORCE DEVELOPMENT
Abstract:
Current structural design practices for infrastructure projects rely on time-intensive manual calculations and code compliance verification, creating bottlenecks in project delivery and potential for human error. This research presents a multi-agent Large Language Model (LLM) framework that automates code-compliant infrastructural design while maintaining interpretability and verifiability which is essential requirements for infrastructure applications.
The framework employs specialized LLM agents coordinated through task distribution: a Task Dispatcher routes design queries to dedicated Design and Evaluation agents, which interface with deterministic calculation tools programmed according to structural design codes. An Expert Consultation agent enables iterative refinement through natural language interaction, supporting the design optimization process. Case studies on reinforced concrete beam design demonstrated exceptional performance: 97% accuracy compared to industry-standard software (SAP2000), 90% time reduction compared to traditional methods, and complete transparency through step-by-step calculations with explicit code references. Statistical validation across 30 design cases showed Mean Absolute Percentage Error below 3% for critical structural parameters.
The framework’s modular architecture enables adaptation to various infrastructure applications by incorporating different design codes and specialized calculation modules. Engineers can specify requirements in natural language while receiving compliant solutions with detailed explanations, facilitating rapid design iteration and supporting workforce development through transparent educational content. This research demonstrates an approach to infrastructure design automation that preserves engineering judgment while eliminating routine calculation tasks. The framework represents a significant step toward preparing the engineering workforce for AI-enhanced infrastructure development.
Jinxin Chen is a PhD candidate in Civil Engineering at Stevens Institute of Technology, specializing in the integration of artificial intelligence and machine learning into structural design and infrastructure applications. His research focuses on developing automated design tools that enhance workforce efficiency while maintaining engineering rigor and code compliance. Mr. Chen has authored multiple peer-reviewed publications on computational methods in structural engineering, including work on machine learning for ultra-high-performance concrete and AI-assisted design frameworks. His current research on large language model frameworks for infrastructure design aims to streamline project delivery while supporting knowledge transfer and workforce development in the engineering profession.
Presentation Slides:
To be added.