This project investigates the extent to which agentic large language models, operating on structured representations of design and coupled with simulation, analysis, and existing design optimisation tools, can generate novel designs for engineered systems. Specifically, the project focuses on whether such methods can propose alternative sustainable solutions to existing systems that not only reduce environmental impact but also enhance human wellbeing (such as safety, user experience, or overall system resilience) across the lifecycle.

Key to this will be:

• investigating different combinations of underlying design representations and design operations to be used by the agentic AI
• coupling these with appropriate analysis, simulation,  optimization tools and human-in-the-loop simulation frameworks capable of evaluating environmental and wellbeing (such as physical and cognitive ergonomics) impact
• selecting relevant case studies in a relevant domain of study and benchmarking against existing methods on performance, sustainability, safety metrics such as embodied carbon, resource depletion, health, and other impact factors

Contemporary engineering design is increasingly shaped by the need to address a broad, often competing set of objectives, with a major focus on sustainability and environmental impact across the full life cycle of engineered systems. Despite these demands, dominant approaches to engineering design continue to emphasise incremental modification and component-level refinement of existing system representations, rather than enabling principled exploration of alternative system architectures under tightly coupled objectives. In response, recent computational approaches have begun to integrate artificial intelligence and machine learning into the design process. Deep generative models have demonstrated effectiveness in exploring high-dimensional design spaces and optimising parameters within predefined system representations. However, these methods remain constrained by fixed parameterisations, limiting their ability to support open-ended synthesis and the generation of novel designs. Although large language models (LLMs) show strong capabilities in code and specification generation, they still currently operate as open-loop components in engineering design contexts, limiting the ability to ensure geometric consistency or physical validity of the output. This project aims to overcome these challenges, integrating agentic AI as an orchestrator of an open-ended design generator combined with analysis and simulation tools to inform the design.

The key contributions expected are:

• a literature review on both the history as well as the state-of-the-art of computational methods for design (design grammars, generative design, data-driven/machine learning-based design)
• application of agentic LLMs to combinations of existing design representations (design grammars, topology optimisation parametrisations) and simulation/analysis tools
• development of generalised design representations that can leverage LLM latent domain knowledge, as well as support for structured interaction with simulation/analysis tools
• selecting relevant case studies in a relevant domain of study and benchmarking against existing methods on performance and sustainability metrics such as embodied carbon, resource depletion or other impact factors