[Local LLM Implementation Case Study] AI Utilization That Balances Confidential Information Protection and Business Efficiency | Matsuo Lab Startup Athena Develops Local LLM for Joyo Bank to Streamline Banking Operations Without Sending Confidential Data Externally

Athena Technologies, Inc. (Headquarters: Bunkyo-ku, Tokyo; CEO: Takeshi Abe; hereinafter "Athena"), a startup originating from the Matsuo Laboratory (*1), announces that it has developed and provided "JOYO AI AGENT," a specialized business agent using a local LLM, for The Joyo Bank, Ltd. (Location: Mito City, Ibaraki Prefecture; President: Tetsuya Akino; hereinafter "Joyo Bank"). This project, following the PoC announced in July 2025 (Press release here), has completed the development of a common platform in a cloud private network environment and the implementation of "Translation & Masking" functionality. Additionally, to pursue further practicality, additional verification of "Automated Business Document Creation" and "Approval Document Review" is currently being trialed. This article presents the initiative centered on these four use cases as a DX case study for banking operations that balances strict security with on-site convenience.

The Journey from PoC to Building "JOYO AI AGENT" Since the PoC launch announced in July 2025, the project has pursued the dual goals of ensuring data security and improving business efficiency through generative AI in banking operations where strict information protection is required. As a result, the practicality of translation and masking functions, as well as the effectiveness of AI utilization in other specialized operations, was confirmed. Based on these verification results, development of "JOYO AI AGENT" proceeded, and a new PoC for additional feature development was initiated to pursue further practicality. ■ Three Design Principles Optimizing AI Implementation for Banking Operations In building "JOYO AI AGENT," a design emphasizing three pillars—security, scalability, and convenience—was adopted to balance the strict security required in banking operations with practical usability in the field. ・[Security] A "Private Network Environment" Unreachable from the Internet That Thoroughly Protects Confidential Information Rather than using external generative AI services, a local LLM was built on a private cloud (Azure VNet) physically isolated from the internet environment. By directing data transmission to a dedicated isolated environment, the risk of leaking confidential and personal information is eliminated, enabling adaptation to banking operations where AI utilization was previously difficult.

・[Scalability] Designing a "Local LLM Operations Platform" to Accelerate Horizontal Deployment As the core of the system, a common local LLM and a "Local LLM Operations Platform" handling login, log management, and other functions were built. Features integrated into "JOYO AI AGENT," such as translation and masking, are all implemented as "apps" that run on top of this common platform. This design means that when introducing AI to new business operations in the future, there is no need to rebuild the entire system from scratch—additional features can be developed quickly and efficiently.