AI agents for structural design in early-stage engineering workflows

Technology
Conceptual
University
Scouted & summarized by Halo AI

ETH Zurich is developing AI agents for early-stage structural design that enable engineers to transform natural-language intent into editable 3D structural models. By bridging the gap between LLM-based reasoning and domain-specific structural data, this research project aims to streamline repetitive modeling tasks, increase design exploration, and enhance collaboration between human experts and AI systems in the AEC industry.

Overview

The AI agents for structural design initiative is a research project developed at ETH Zurich, aimed at transforming early-stage engineering workflows. By utilizing large language models (LLMs) and agentic frameworks, the system enables structural engineers to translate natural-language design intent into editable, domain-specific 3D structural models. This approach seeks to reduce the manual, fragmented nature of current design processes, allowing teams to explore a wider range of structural alternatives efficiently while maintaining human control over critical engineering decisions.

Technical specifications
  • Workflow: The system connects a specialized AI agent to a structural modeling API via a Model Context Protocol (MCP) server.
  • Data structure: It employs a COMPAS-based graph data structure to represent building elements such as joints, frames, slabs, and walls with semantic richness, allowing for inspection and modification.
  • Interaction: Designers provide natural-language prompts to describe structural models or request modifications, which the agent translates into deterministic modeling actions.
  • Integration: The output is an editable, serialized artifact that can be integrated into downstream architecture, engineering, and construction (AEC) workflows.
Technology readiness level

The project is currently at a conceptual research stage, specifically rated at TRL 3. It has demonstrated a functional proof-of-concept workflow that integrates language-model reasoning with deterministic structural modeling operations. While it successfully generates editable graph-based artifacts, it is currently in the development and evaluation phase, with published studies noting that current prototypes require further research regarding model size, analysis readiness, and handling of complex spatial reasoning.

Generated Nov 17, 2026. Please verify key information.

About ETH Zurich

ETH Zurich is a comprehensive, STEM‑focused public research university within Switzerland’s ETH Domain, recognized for large‑scale, interdisciplinary science and engineering. Industry engages on the Hönggerberg “Science City” campus through co‑located laboratories, prototyping spaces, and collaboration suites, and via a joint translational center with the University of Zurich that links researchers to clinical partners. A Basel campus positions faculty and students alongside one of Europe’s densest pharma‑biotech ecosystems, streamlining sponsored research and talent pipelines. Research is supported by the Swiss National Science Foundation, Innosuisse, and competitive European programs. ETH transfer provides IP management, licensing, and startup support to accelerate commercialization.

Halo home
Partner smarter. Move faster.
Get new partnering requests
delivered to your inbox.