[Digital Today reporter Choi Jae-won (최재원)] Japan's telecoms company NTT has launched a test to introduce its self-developed AI into administrative work at elementary and junior high schools. It aims to reduce staff burdens from document work and increase time for communicating with students. The test will also verify the potential use of homegrown AI in education settings.

On Oct. 14 (local time), Japanese media outlet ITmedia reported that NTT West, working with Ibaraki city in Osaka prefecture, began a proof-of-concept project the previous day at some municipal elementary and junior high schools using the large language model Tsuzumi 2. The project stems from the problem that school staff spend significant time searching for work materials scattered across multiple servers and files and writing documents.

Staff can ask questions in a dedicated AI application, as if chatting, to search for needed school administrative documents or summarise content. It also supports drafting various documents, including official documents and work reports.

Data security is central. The system adopts an on-premises approach, running on internal servers rather than external cloud services. It is also designed to reduce the risk that sensitive information about students and staff is improperly exposed by limiting access rights by user.

Tsuzumi 2 is a Japanese-made LLM that NTT developed independently rather than based on another company's AI model. It has a lightweight architecture that can run on a single GPU, enabling operation without expensive AI servers. It is also used in fields handling confidential information, including finance, healthcare and public administration, due to strengths in understanding Japanese documents, extracting information and summarising.

The test runs until March 31, 2027. The two organisations plan to assess document search and drafting time, the usefulness of AI responses, ease of use and willingness to keep using the system through staff surveys and interviews. NTT West explained, "This is Japan's first elementary and junior high school administrative AI test using a domestic private LLM."

The case allows schools to weigh both cost and security burdens when seeking to run AI on internal facilities rather than using external cloud services. It is also expected to serve as a test bed to confirm the level of efficiency from search accuracy and document drafting support in real staff work, beyond a simple functional experiment.