English 
NGK and ラボAI have worked together to build a new way of developing large-scale simulation software using generative AI, and they are already putting it to use in NGK’s organic compound crystal discovery service.
The problem they are tackling is not small. The pharmaceutical industry experiences changes in crystal structure because temperature and concentration and infrared irradiation cause new conditions. The changes that occur during this process directly affect both solubility and stability. The testing process requires experimental work because all possible combinations need to be tested. The process requires excessive time together with significant effort.
This is where simulation comes in, but even that has its own bottleneck. Traditional simulation software development is slow and heavily manual. Writing and validating complex mathematical models as code takes time and carries a real risk of errors.
こちらもお読みください： 地球温暖化防止に役立つ日本の新しい水力発電技術
The joint project focused on changing that process. Instead of building everything line by line, they fed mathematical formulas, constraints, and conditions into a generative AI system in a structured format. The goal was to see if the AI could turn that into working simulation code without cutting corners or misinterpreting the logic.
The result was clear. With the right inputs, including defined development rules and base code, the AI was able to generate accurate programs. More importantly, the workload shifted. A general-purpose coding agent could handle most of the implementation, bringing human effort down to roughly one third of what it used to be.
That changes where engineers spend their time. Less on repetitive coding, more on validating theories, refining models, and interpreting results.
日本ガイシ plans to apply this approach directly to its crystal form prediction software, with the aim of improving speed and flexibility in responding to customer needs. Both companies also see this going beyond one use case, with plans to expand the method into other simulation domains and shorten the path from research to real-world application.
日本語