West Japan Railway Company used generative AI to automate API definition creation and review. In a PoC, the AI detected 87.5% of errors and addressed all 9 remediation instructions, confirming a substantial reduction in work time.
West Japan Railway Company established a common API platform in 2020 and developed a framework to meet growing demand for the external sharing of railway operating information and other data. The company adopted a policy of avoiding one-to-one connections between multiple business systems and consolidating external access through a single gateway, with the aim of facilitating external integration.
To address these challenges, the company explored the use of generative AI and introduced, as a PoC, an API definition creation support tool proposed by TISI. It tested a mechanism designed to reduce the workload of both the engineering team and the platform operations team.
To improve AI accuracy, the system was designed to ensure that the generative AI appropriately references information such as API specifications, authentication rules and documentation conventions. Testing was also conducted in an environment that met the company’s internal security standards. An automated code-generation function from scratch was also incorporated.
As a result of the PoC, the proportion of errors in code created by engineers that were flagged by the AI reached 87.5%, and all 9 remediation instructions were addressed. Work that had previously taken 1–2 hours for remediation and more than 3 hours for review was reduced to approximately 30 minutes and several minutes to several tens of minutes, respectively.
Going forward, the company plans to move the system into production operations and expand integration through the common API platform to approximately 20 business systems. It also plans to broaden the scope of application, including promoting data integration to meet rising MaaS demand, such as shuttle bus integration for the Osaka-Kansai Expo, and automating requirements definition documents and design documents.