Proceedings of MATSUS Fall 2026 Conference (MATSUSFall26)
Publication date: 22nd July 2026
Automated, high-throughput laboratories now let us assemble and cycle battery cells reproducibly and at scale, reducing the cell-to-cell variability of manual fabrication but opening a growing gap between the amount of data a lab can produce and what researchers can curate, process, and interpret using traditional workflows. Large language models (LLMs) have improved rapidly in recent years and offer a promising way to close this gap; yet their inner workings remain only partially understood, which matters for how far we can trust them in a scientific setting.
Recent work from our group shows that LLMs can have internal structure to how they store scientific information [1] and the way information is presented to them can have a significant impact on their outputs and associated confidence [2]. Building on this, the work I present is organised around three threads reflected in the title: keeping experimental data FAIR, making automated labs and research consortia aware of their own operation, and grounding both in a clearer understanding of the models themselves.
Building on this, the work I present is organised around three threads: ensuring data follows the FAIR principles, making automated labs and research consortia aware of their own operation, and grounding both in a clearer understanding of the models themselves.
