L-SCOPE: An LLM-Assisted Interactive Platform for Efficient Building Design Optimization

Abstract

The building and construction sector accounts for a significant share of global energy use and greenhouse gas emissions, highlighting the importance of focusing on sustainable design for climate change mitigation. Early-stage design decisions are crucial in shaping a building’s life cycle energy performance. However, sustainable building design optimization often relies on simulation-based optimization (SBO), of which complex objectives and numerous design parameters result in high computational complexity and limited user understanding of how individual parameters influence the outcome. This study introduces an LLM-assisted interactive platform to enhance the efficiency and interpretability of SBO. By integrating the large language model (LLM) with an information gain-guided framework, the platform facilitates an end-to-end pipeline—from decomposing analytical tasks, and generating code, to user-triggered execution, and visual result interpretation. Information gain analysis enables design complexity reduction by narrowing value ranges of influential parameters and fixing less impactful ones, ultimately enhancing optimization efficiency. The prompt-driven interface allows non-expert users to engage directly with the analysis process, therefore empowering users with interpretable, stakeholder-oriented insights and AI-assisted decision support. A case study on sustainable building design demonstrates that the platform improves both the interpretability and performance of the optimization process compared to conventional SBO methods, while operating under the same computational budget. The contributions of this study are twofold: (i) a novel approach that combines LLM capabilities with information-theoretic analysis to advance interpretable and efficient building design optimization; and (ii) a novel language-based interface that facilitates active participation and decision-making by non-expert stakeholders.