Advancing materials science with LLMs (MatSci-LLM)

The MatSci-LLM project explores how large language models (LLMs) can become more useful for materials scientists. By adapting AI to specialised terminology and complex scientific data, we are helping researchers streamline information analysis, answer complex questions, and identify new directions for future research.

Close-up of a scientist holding vials of liquid. — Photo: Julia Koblitz / Unsplash.

Materials scientists work with vast amounts of specialised information, including scientific papers, experimental data, phase diagrams, tables and crystal structures. Reviewing and connecting this information is time-consuming and leaves less time for experiments and scientific discovery.

Although current AI systems are effective at processing text, the models still struggle with various types of structured data and scientific visualisations used in materials science.

MatSci-LLM investigates how AI can support researchers at different stages of their work, from identifying patterns and using calculation tools to proposing hypotheses for further investigation. The project will also contribute to the development of AI research assistants and agent-based systems for scientific tasks, giving scientists better tools for investigating questions and testing new ideas.

How we are applying LLMs to materials science

As a multidisciplinary project, MatSci-LLM brings together expertise in materials science, natural language processing and image analysis to support researchers at different stages of the research process.

The project will:

adapting language and multimodal AI models to understand materials science data, including phase diagrams, crystal structures and other scientific visualisations
developing new ways of representing material structures so they can be processed more effectively by the models
combining LLMs with scientific knowledge and calculation tools to improve their reasoning about materials science problems
developing AI-powered research assistants and agent-based systems that can propose new hypotheses to be tested through simulations and laboratory experiments.

By transforming how materials data is processed and understood, MatSci-LLM bridges the gap between advanced AI and materials science, laying the groundwork to redefine the speed and scope of scientific discovery.