Xinwei Zhuang

(pronounced as Sheen-way Juin , 庄新伟)

I am a 5th year PhD candidate in Architecture. I am honored to be advised by Prof. Luisa Caldas, coadvised by Prof. Simon Schleicher in the Building Science, Technology and Sustainability program at UC Berkeley. I'm a member of XR lab and Center for the Built Environment (CBE), and President at Immersive Design Student Club. I worked as both a researcher and an architect, and served as a research assistant at the division of Building Technology and Urban Systems, Lawrence Berkeley National Lab, focusing on using Scout to estimate energy and carbon impact of energy conservation measurements.

Email  /  CV  /  Bio  /  Scholar  /  LinkedIn

Selected Research

My research is centered on machine-learning-aided generative design, with a special emphasis on energy efficiency and resilience. My recent work concentrates on building stock analysis through generative models and modeling urban environments with graph theory. I am actively seeking collaboration opportunities in these areas.

We provide a open-source algorithm to compute urban-scale solar potential. Details coming soon.

Zhuang X., Lyv G., Zhao Z. and Caldas, L. (2025) Rapid Assessment of Solar Potential for Building Surfaces in Complex Urban Morphologies Based on Vector Processing, Solar Energy

This literature review examines the critical role of data encoding, such as images, point clouds, topology graphs, and textual descriptions, in revolutionizing architectural design through machine learning. By using different data formats, architects can move beyond rule-based methods and unlock unprecedented creative possibilities in architecture. The paper emphasizes both significant advancements and ongoing challenges, providing insights into the future AI-driven architectural design and research.

Zhuang X., Zhu P., Yang, A. and Caldas, L. (2025) Machine learning for generative architectural design: Advancements, opportunities, and challenges, Automation in Construction, Volume 174, 2025, 106129, ISSN 0926-5805, doi: 10.1016/j.autcon.2025.106129.

We propose Hierarchical Urban Graph, a multi-layered network approach designed to optimize urban neighborhood partitions for energy autonomy and resilience by balancing renewable energy generation and consumption. Through a case study of San Francisco, the methodology demonstrates improved stability in energy end-use patterns and reduces dependency on centralized grids. Ultimately, this scalable, data-driven framework provides a tool for sustainable urban energy planning and infrastructure design.

Zhuang X., Chu X., Liang, J., Zhu P., Gonzalez, M. and Caldas, L. (2024) Across scales: Hierarchical Urban Graph for Neighborhoods Partition and Decentralized Energy Autonomy, 2024 Conference on Association for Computer Aided Design in Architecture (ACADIA): Designing Change, Calgary, Canada.

Refined and open-source release of an automated, geometrically detailed, localized building archetype generator for energy modeling. This paper delves into the specifics of an algorithm designed to automatically generate building archetypes tailored to local contexts, with a focus on detailed geometric resolution for energy modeling purposes. We provide all codes and data necessary for replication.

Zhuang X., Huang Z., Zeng W. and Caldas L. (2023) MARL: Multi-scale Archetype Representation Learning for Urban Building Energy Modeling, Computer Vision Aided Architectural Design (CVAAD), International Conference on Computer Vision (ICCV), Paris, France

Paper presents a high-level overview of the algorithm behind the automated generation of localized building archetypes, outlines the algorithm's value proposition and its practical applications.

Zhuang X., Huang Z., Zeng W. and Caldas L. (2023) Encoding Urban Ecologies: Automated Building Archetype Generation through Self-Supervised Learning for Energy Modeling, 2023 Conference on Association for Computer Aided Design in Architecture (ACADIA): Habits of the Anthropocene, Denver, United States of America

Introduces 3D learning in the architectural domain, leveraging generative models to understand and manipulate the complex morphology of buildings. By constructing an original database specifically tailored for this purpose, the research offers capabilities for high-dimensional design exploration and energy performance evaluation.

Zhuang X., Ju Y., Yang, A. and Caldas L. (2023) Synthesis and Generation for 3D Architecture Volume with Generative Modeling, International Journal of Architecture Computing, Vol. 21, Issue 2: AI, Architecture, Accessibility, & Data Justice, doi.10.1177/14780771231168233