Our research focuses on developing AI-driven methodologies for materials discovery and design. 

This includes (1) automated literature data collection using language models, (2) ontology-based structuring of materials data, (3) AI models for analyzing and predicting structure-property relationships, (4) generative AI-driven exploration of novel materials, and (5) machine learning potential-based simulations for material synthesis, integrating a full-cycle data-AI approach to materials design.

Our research focuses on graph-based recognition of crystal structures to enable AI-driven materials discovery. Leveraging large language models (LLMs) and generative AI techniques such as diffusion models and flow matching, we aim to explore stable, unique, and novel (S.U.N.) materials. By integrating graph representations with generative AI, we seek to accelerate the identification and design of breakthrough materials with enhanced stability and functionality.

Our research focuses on high-speed, high-accuracy fine-tuning of universal potentials (CHGNet_0.4.0, MACE-MPA-0, NequIP, and Allegro) using graph-based structural representations. Leveraging these advanced machine learning potentials, we aim to screen material properties efficiently and simulate synthesis processes across multiple scales, from the atomic level to the microscale. This approach enables precise multiscale simulations, accelerating the discovery and design of novel materials.

We conduct research to elucidate reaction mechanisms and understand the principles of material property manifestation through in-depth, multi-faceted analyses—including thermodynamic, kinetic, electrochemical, and electronic structure investigations—based on first-principles density functional theory (DFT) calculations. Our study integrates computational materials science approaches to provide fundamental insights into material behavior and performance.