[Local LLM Case Study] AI Utilization Balancing Confidential Information Protection and Operational Efficiency | Matsuo Lab Startup Athena Streamlines Banking Operations for Joyo Bank with Local LLM that Prevents External Transmission of Sensitive Data

Athena Technologies Inc. (Headquarters: Bunkyo-ku, Tokyo; CEO: Takeshi Abe; hereinafter "Athena"), a Matsuo Lab startup (*1), is pleased to announce that it has developed and provided "JOYO AI AGENT," a business-specific agent utilizing local LLM, to Joyo Bank, Ltd. (Location: Mito City, Ibaraki Prefecture; President: Tetsuya Akino; hereinafter "Joyo Bank").

Following the Proof of Concept (PoC) announced in July 2025 (click here for the press release), this project has completed the development of a common platform in a cloud closed-network environment and the implementation of "translation and masking" functions. Furthermore, in pursuit of greater practicality, additional verification regarding "automated business document creation" and "credit review" is currently being trialed.

In this article, we will disclose these initiatives centered on four use cases as examples of DX in banking operations that achieve both strict security and on-site convenience.

Journey from PoC to the Construction of "JOYO AI AGENT"

Since the start of the PoC announced in July 2025, we have pursued the compatibility of data security and operational 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 tasks, were confirmed.

Based on these verification results, we proceeded with the development of "JOYO AI AGENT" and initiated a PoC for additional functional development in pursuit of further practicality.

Three Design Principles for Optimizing AI Implementation in Banking Operations

In building "JOYO AI AGENT," we adopted a design that emphasizes three pillars—safety, scalability, and convenience—to balance the strict security required for banking operations with practical utility in the field.

・ [Safety] A "Closed Network Environment" with No Internet Access to Protect Confidential Information Instead of using external generative AI services, we built a local LLM on a closed cloud (Azure VNet) physically isolated from the internet environment. By setting the data destination to a dedicated isolated environment, we eliminate the risk of leaking confidential or personal information, enabling adaptation to banking operations where AI utilization was previously difficult.

・ [Scalability] Designing a "Local LLM Operation Platform" to Accelerate Horizontal Expansion As the core of the system, we built a common local LLM and a "Local LLM Operation Platform" responsible for login and log management. Functions equipped in "JOYO AI AGENT," such as translation and masking, are all implemented in the form of "apps" running on this common platform. With this design, when introducing AI to new operations in the future, it is not necessary to rebuild the entire system from scratch, and additional functions can be developed speedily and efficiently.