WS6
AI-Collaborative Computational Design: Customized Tool Making with Large Language Models
Learning Objectives
Participants will familiarize themselves with the cutting edge of generative AI applications in computational design workflows and themes. The workshop will enable participants to adapt Raven, a generative AI plugin for Grasshopper, to work with their own scripts and solve for specific workflows. Small teams will explore and develop customized instances that are tailored to solve and support computational design tasks with simulation informed feedback, such as iterative development of building models with energy simulation feedback, physics informed geometry etc. The participants will leave the workshop with a full understanding of the capabilities and limitations of generative ai tools in Grasshopper, and the ability to modify and customize these tools for their own practice and workflows.
Workshop description
Parametric design has become synonymous with computational design practice, driven by its broad applicabilities and the enthusiastic and generous community fostering a vibrant and rapidly expanding ecosystem. Grasshopper, specifically, has grown to be the preeminent platform for computational design, characterized by its openness and extensive community contributions. Its ecosystem includes a multitude of interoperable plugins enabling workflows that range from complex geometry generation and robotic tool path planning to sophisticated analyses like energy simulations, daylight studies, structural assessments, and physics-based simulations. The interoperability inherent in Grasshopper's directed node-graph interface continuously expands its functionalities with every new plugin added, creating a powerful network of possibilities.
We propose that Grasshopper, traditionally thought of mostly as a visual interface for human-CAD communication, is ideally positioned to become the symbolic language bridging human designers, artificial intelligence, and CAD software. Grasshopper abstracts functions into clearly defined nodes with minimal parameters, abstracting intricate computational processes that happen within the Rhino kernel. This abstraction positions Grasshopper as an optimal symbolic interface, uniquely suited for interaction with AI systems that excel at manipulating abstract representations - inviting comparisons to the successes seen in neural-symbolic AI models on frontier math and geometry proof solving.
Our workshop explores precisely this potential. Participants will leverage Raven, a novel AI plugin for Grasshopper, that uses Large Language Models (LLM), to dynamically construct and refine parametric workflows at various abstraction levels—from individual node-level customization to high-level workflow orchestration. We show how the generalized LLM can be customized to incorporate highly specific user workflows and make them more accessible. As a case study we will explore how the new workflow can be used to engage with graph-based generation of residential building floor plans. The LLM model allows for users to create and adapt complex graph structures that are used to create building layouts. Together with the workshop participants we will explore how the AI driven tools can be adapted to incorporate more custom workflows - from spatial analysis to environmental simulations.
This workshop aims to demonstrate how integrating symbolic AI into Grasshopper workflows can drastically expand the accessibility and functional scope of computational design, lowering barriers to sophisticated analyses and fostering deeper human-AI collaboration.
Participant Prerequisites
Keywords: Generative AI, Advanced Architectural Geometry, Parametric Design, Neurosymbolic Systems
Required Skill: Advanced knowledge of Grasshopper and Rhino, familiarity with design scripting, visual programming, and computational methods. Proficiency in Python is highly recommended.
Required Software: Rhino 8 (not 7 or lower), Grasshopper, Python (within Grasshopper), custom AI–Grasshopper plugin (provided by instructors).
Required Hardware: None other than personal computers
Workshop Information
Workshop Leaders
Moritz Rietschel, UC Berkeley
Moritz Rietschel is a designer and researcher from Switzerland. He completed his undergraduate studies in Austria, where he worked on robotic fabrication and experimental design methods in the Creative Robotics Research Group led by Professor Johannes Braumann. He came to UC Berkeley to join Professor Simon Schleicher’s robotic fabrication lab and to explore human–AI collaboration with Professor Kyle Steinfeld. His current work focuses on how large language models can engage in computational design reasoning—specifically in generating and adapting parametric systems within architectural workflows. He is a co-author of Mediating Modes of Thought (ACADIA 2024) and is actively publishing further research in the emerging field of AI–human co-design, with a focus on systems capable of geometric and algorithmic reasoning.
Ramon Weber, UC Berkeley
Ramon Weber is an Assistant Professor of Architecture at the College of Environmental Design at UC Berkeley, where he leads the Spatial Systems Lab. Working at the interface of sustainable construction, digital structural design, building technology and computing, he investigates how computational design methods and simulation tools can create more sustainable architecture. He completed his PhD at MIT in Building Technology, where he developed frameworks and automation tools for the design and analysis of low-carbon buildings. He is a graduate from ETH Zurich, the University of Stuttgart, and MIT’s Media Lab where he was a researcher at the Mediated Matter Group. He previously worked for Zaha Hadid Architects in London, and he was involved in projects across scales at the ZHA|CODE research group. His personal and professional work has been published and presented internationally in both scientific and design venues such as the Venice Architecture Biennale, SFMOMA, and Ars Electronica, as well as the Journal Nature Communications, Solar Energy, Building and Environment, 3D Printing and Additive Manufacturing, and ACADIA