Text2Structures: Communicative Agents for Early Stage Structural Design

Abstract: Communicative agents powered by Large Language Models have potential to transform structural engineering design. This study introduces Text2Structures, a multi-agent system for early conceptual design. Using Retrieval Augmented Generation, agents generate a set of structural design recommendations. The demonstration of a simple parking structure design verifies that agents can recommend materials, structural systems, and dimensions for floors and beams. Using Text2Structures, users can input brief requirements in natural language, integrate regulatory and prescriptive documents from local jurisdictions, and receive transparent and justified structural recommendations. Although further refinements are needed, this study lays a foundation for communicative agents in structural design.