From natural-language description to formalization: Capturing human-centered social design intentions in BIM using large language models

The role of large language models (LLMs) in the architecture, engineering, and construction (AEC) sector has been increasing rapidly over the past few years, as they are being used as assistants, analyzers, and chat agents to optimize building designs through their context-based text-generation capabilities. In this paper, we present ProSpect, a software tool developed to capture architects’ human-centered social design intentions (SDIs) by integrating building information modeling (BIM) and LLMs. The aim is to enable architects to clearly and explicitly integrate qualitative design intentions into BIM models. This work builds on our previously developed formalization framework, ProFormalize, which provides a domain-specific language to capture design intentions that particularly elicit human-centered criteria (e.g., curiosity and comfort). We present the development of ProSpect, following a co-creation approach and a case study-driven methodology that aim to empirically assess the validity of our framework and the usability of our software tool, comparing its LLM-based (latest) version with its wizard-based (previous) version. Our study shows that the LLM-based solution is more efficient at capturing and representing SDIs, achieving approximately 9% higher accuracy on trained prompts than on untrained ones.