東洋紡 introduced its initiative at AWS Summit Japan 2026 to draw out manufacturing knowledge using an AI agent “taught and nurtured” with language from the frontline, as part of TX (Toyobo Transformation). The company discussed the design and operation of an agentic RAG system developed together with frontline personnel.
東洋紡 is introducing generative AI technology to the frontline as part of TX. Sakakura Hiroya (坂倉広也) spoke on TX and discussed the design and operation of an agentic RAG system developed together with frontline personnel.
A common problem emerged: frontline knowledge could not be found, communicated or put to use. The presentation pointed to the three conditions of workload pressures, dependence on individuals and increasing complexity as obstacles to sharing frontline knowledge. The goal was to connect the language used on the frontline with knowledge contained in company documents.
In the initial phase, a simple RAG system was used to build a knowledge base from frontline terminology and internal documents. However, accuracy remained at approximately 0.45, prompting comments from the frontline that the system was unusable. The operational burden on frontline personnel and the prospects for achieving tangible benefits were identified as challenges.