LapidaryEngine Enables Conversational AI for Crystal Material Design

The vision of designing bespoke crystal materials using natural language instructions has been inspired by the advent of Large Language Models (LLMs). While existing text-to-crystal generative models have made initial progress, they are often limited by requiring highly structured input formats and offering only one-directional generation, preventing interactive refinement. LapidaryEngine has been developed to enable fully conversational crystal generation. This model accepts free-form natural language requests and facilitates iterative refinement and editing of crystal structures through a dialogue-like interaction. A key innovation is the use of a "pivot representation," an intermediate form that allows for bidirectional translation between text and crystal structures, even without direct paired datasets. This pivot representation enables robust interpretation of user feedback and precise structural control during the design process. LapidaryEngine's capabilities have been demonstrated across various tasks, including discovering insulators, optimizing stability, modifying composition, and editing structures, showcasing its ability to align generated materials with user intent interactively.