This research investigates the role of large language models (LLMs) as computational design agents within Rhino 8 through Python-mediated workflows. The central premise is that LLMs can function as translation mechanisms between architectural intent, expressed in natural language, and executable geometric operations, encoded as Python scripts. Rather than treating LLMs as autonomous designers, the study frames them as intermediaries within a structured human–machine pipeline: human-defined intent, prompt construction, LLM-based code generation, geometric execution in Rhino, and iterative evaluation.

The project develops a systematic framework for integrating LLMs into computational design processes, focusing on reproducibility, control, and interpretability. It defines prompt architectures for common architectural tasks, including procedural massing, spatial organization, patterning, and agent-based simulations, and evaluates the reliability of LLM-generated scripts against criteria such as Python correctness, Rhino API compatibility, geometric validity, and architectural usefulness.

By positioning LLMs as operational components within computational design workflows rather than as creative substitutes, this work contributes a technical and conceptual foundation for integrating language-based models into architectural practice, pedagogy, and research.