ArchiJury: Exploring the Capabilities of Vision-Language Models to Generate Architectural Critique

Artificial Intelligence (AI) offers a potent opportunity to rethink architectural critique, in cases such as architectural design competitions. The challenge lies in capturing the interpretive depth required for design evaluation—an inherently human process that connects intuition, reasoning, and contextual sensitivity. Building on this premise, the proposed approach uses a domain-specific dataset, curated and validated by experienced architects as domain experts, to train a context-aware Visual-Language Model (VLM) capable of delivering a nuanced critique. The model development follows two distinct phases: an initial prototype (v1) explores feasibility through classification of visual architectural attributes, while the second phase (v2) evolves into a structure generating detailed critique texts guided by predefined criteria such as context, form, and programmatic considerations. The proposed model aims to bridge the gap between computational precision and the complexities of architectural judgment, offering a structured yet adaptable framework for utilizing AI in the evaluative aspects of design.By integrating ecological intelligence into this framework, the critique can also assess designs based on their environmental impact and sustainability practices, encouraging a holistic approach that aligns architectural innovation with ecological responsibility. Although still in its early stages, this work opens a pathway to complement traditional review processes with reliable, scalable, and context-sensitive feedback, laying a foundation for incorporating the patterns of tacit knowledge in architectural design into the review process.