A demo implementation of DUCTILE
DUCTILE: Agentic Large Language Model Orchestration of Engineering Analysis in Product Development Design Practice Pradas-Gómez, Brahma and Isaksson, J. Mech. Des. 2026
The purpose is to highlight how engineers can interact with tools and agents, and how these agents can orchestrate the update of pre-determined automations. This example agent lacks access to design practices, so it relies on the knowledge and capabilities of LLMs to do an acceptable job. Try different models to see how they affect its performance.
The scenario represents a design investigation and the early sizing of a design. Two problem definitions are given, or you can define your own: describe the geometry, material, supports and loads, and the agent extracts a problem definition from your description.
Beta: the FE Agent is under active development and may be updated while you use it. You will see a notice when that happens; the page reloads by itself and your conversation is usually kept.
Privacy: your messages go to the language model you choose in the menu: Anthropic's API (Claude), OpenAI's API (only with your own key), or our own server at Chalmers for the free models. Once research logging is enabled, conversations and runs are recorded as described above, pseudonymised; until then they are kept only in memory while you use the app. Usage and cost are logged per access code. Runs are deleted an hour after they finish.