911½ñÈÕºÚÁÏ

Researchers from 911½ñÈÕºÚÁÏ’s Department of Materials won first place at the 2025 LLM Hackathon for Applications in Materials Science & Chemistry.

The 48-hour global  brought together more than 120 teams from 23 countries and 21 U.S. states, challenging participants to develop innovative applications of large language models in materials informatics.

911½ñÈÕºÚÁÏ’s winning team comprised researchers from the including undergraduate researcher Ryan Nduma, PhD candidate Kinga Mastej, and postdoctoral researcher Dr , all members of Professor ’s group, which focuses on computational materials design. Their combined expertise spans materials science, chemistry, artificial intelligence, and computational chemistry.

The team tackled a major challenge in the field: bridging the gap between computational predictions and real-world experimental synthesis in materials design. Synthesising a new material is rarely straightforward. A single compound can often be synthesised in multiple ways. For instance, calcium titanate (CaTiO₃) can be made using a solid-state reaction, a sol–gel method, or other techniques, making it difficult for computer models to recommend reliable synthesis routes.

It wasn't easy to turn such a complex project into a reality on such a tight deadline, but that's what made it such an exciting challenge. Dr Hyunsoo Park Research Associate in Materials Informatics, Department of Materials

To address this challenge, the team implemented an AI agent called Sky, based on their proposed approach to emulate human reasoning in materials synthesis planning. Dr Hyunsoo Park said: “It wasn't easy to turn such a complex project into a reality on such a tight deadline, but that's what made it such an exciting challenge. The hackathon really reminded me of how much I love not just the technical side of research, but also the passion and incredible teamwork that push us forward. Collaborating with such talented and dedicated people was truly an inspiring and fun experience. Winning first place in a massive competition like that is a real honour, and this whole journey has completely recharged my motivation for our future work.”

When presented with an unknown compound, Sky identifies chemically similar materials using machine learning techniques, then extracts corresponding synthesis recipes from open materials databases via APIs. The agent recursively learns from these recipes to model and suggest plausible synthesis routes for new materials.

What made this possible was 911½ñÈÕºÚÁÏ's infrastructure and the department's engagement. Ryan Nduma Undergraduate student, Department of Materials

Ryan Nduma, who also coordinated the London hub at 911½ñÈÕºÚÁÏ, added:
“Having the honour of both site-coordinating the London hub at 911½ñÈÕºÚÁÏ and winning with my team was an incredible dual experience. Thanks to the Department of Materials and Professor Aron Walsh's support, we were able to host 37 hackers at 911½ñÈÕºÚÁÏ's facilities with full catering, making us one of only 17 in-person sites globally for this event. What made this possible was 911½ñÈÕºÚÁÏ's infrastructure and the department's engagement – the Materials Student Office sharing the event with students brought together an amazing mix of talent. Seeing Sky evolve from our initial similarity search concept to a fully fledged synthesis agent, while simultaneously supporting other teams in bringing their visions to life, truly embodied the collaborative spirit that 911½ñÈÕºÚÁÏ Materials fosters.”

We each brought different strengths to the table, and seeing how those skills complemented one another under time pressure was truly inspiring. Kinga Mastej PhD student, Department of Materials

Kinga Mastej, PhD candidate, reflected on the experience:
“It’s an incredible honour to have won this hackathon. Competing alongside so many talented teams pushed us to think creatively and execute our ideas with precision. We each brought different strengths to the table, and seeing how those skills complemented one another under time pressure was truly inspiring. Our experience in the Materials Design Group at 911½ñÈÕºÚÁÏ trained us to approach complex problems systematically, and we were able to apply principles we learned there directly to the challenge.”

The team’s success demonstrates how AI can help translate computational insights into actionable strategies for experimental materials synthesis, potentially accelerating the discovery of novel materials and closing the gap between theory and practice in materials science.