Platform connects directly to internal work systems, automatically generating purchase data when quote documents are attached

A diagram illustrating the key functions of the KRISS AI platform. [KRISS]
A diagram illustrating the key functions of the KRISS AI platform. [KRISS]

South Korea's national metrology institute has built an AI platform designed to reduce researchers' administrative workload and let them focus on their work.

The Korea Research Institute of Standards and Science (KRISS) has developed and launched its own AI platform, called KRISS AI. The system runs entirely on the institute's internal network to address the security constraints that prevent government-funded research institutions from adopting external generative AI services.

KRISS built the air-gapped, intranet-based platform to relieve researchers of complex paperwork and the burden of checking numerous regulations, eliminating concerns about data leaks. The platform integrates with the institute's electronic approval process and establishes a research administration foundation that handles internal knowledge asset utilization and research support simultaneously.

One of the clearest examples of how the platform's AI agent function streamlines work is the small-value direct purchasing process. When a user attaches a quote document — in PDF, Word, or hangul (HWP) format — to the chat window, the AI uses key-value extraction technology to automatically identify and match data such as item names, quantities and amounts.

That data feeds directly into the institute's electronic approval system, auto-filling each field on the application screen and carrying the process through to the approval request stage. Researchers no longer need to manually enter information for multiple items one by one. Leave requests and seminar registrations can also be submitted by typing in plain language in the chat window, with the AI agent connecting to the internal approval system to complete the process.

On the knowledge retrieval side, the platform incorporates retrieval-augmented generation (RAG) to improve the accuracy of internal information searches and knowledge asset use. RAG prevents hallucinations — instances where an AI presents fabricated information from unclear sources as fact. The system identifies each user's individually built database to deliver personalized answers, and lets users compare and search both internal regulations and the higher-level national statutes linked to them in a single view. The platform currently holds approximately 460,000 research reports alongside internal regulations and legal compilations, and supports advanced functions such as report summarization and draft writing.

On the computing infrastructure side, KRISS has put in place a resource virtualization framework to consolidate and manage the high-performance GPU servers currently scattered across the institute, providing the foundation for large-scale research data processing and AI research and development. Resource virtualization allows multiple users to share computing resources simultaneously and efficiently, as if each had their own physically independent AI computer.

KRISS AI is currently running as an open beta service for all staff. Going forward, the institute plans to focus on accumulating internal data and building an intelligent digital work environment essential for more advanced AI use. Plans include progressively linking the platform with document centralization and electronic contract systems, as well as expanding AI coverage across research and administration through personal email integration and automated generation of digital calibration certificates (DCC).

"The KRISS AI platform achieves digital transformation of the work environment while maintaining data security," KRISS President Lee Ho-seong said. "We will continue to drive innovation in the research administration system and create a smart environment that allows researchers to immerse themselves more fully in their work."


nbgkoo@heraldcorp.com