Last year, I went to bed at 2 a.m. and woke up at 6 a.m. to review the national research and development (R&D) budget. This year, my bedtime has moved up by two hours.
At a roundtable on the introduction of artificial intelligence (AI) for government budget review held by the Ministry of Science and ICT on the 14th at the Sejong Government Convention Center, Im Tae-beom, standing member of the Presidential Advisory Council on Science & Technology and head of the Intelligence Research Division at the Korea Electronics Technology Institute (KETI), shared these impressions after actually using an AI specialized for budget review.
National R&D budget review is the process of examining each ministry's submitted R&D project plans and adjusting how much to allocate to which projects for the following year. Every May–June, 166 standing members across 10 technology fields under the operating committee of the Presidential Advisory Council on Science & Technology and budget review officials handle this work.
The problem is that the number of projects and documents to review is increasing. According to the Ministry of Science and ICT, the number of national R&D projects has more than doubled over the past 10 years. The standing members must assess the necessity and budget size of more than 1,000 projects while also checking for projects that are similar or overlapping. Administrative tasks such as summarizing minutes, drafting review opinions, and compiling adjustment results are also substantial.
Starting this year, the Ministry of Science and ICT is using an AI specialized for budget review in the national R&D budget allocation and adjustment process. Its name is "Yeon.Ye.In," short for "research and development budget review artificial intelligence."
The system was developed based on Solar Open from domestic AI company Upstage. The Ministry of Science and ICT said it chose Upstage's model after comprehensively considering currently available graphics processing unit (GPU) resources, development timeline, and feasibility of actual service among several independent AI foundation model project candidates.
The Ministry of Science and ICT, the Korea Institute of Science and Technology Information (KISTI), and the Electronics and Telecommunications Research Institute (ETRI) took part in the development. The system was developed over about five months starting in December 2025 and was built without separate government funding by using personnel and GPU resources owned by KISTI and ETRI.
The AI being introduced now plays a role in assisting with repetitive and time-consuming tasks. It does not directly set budgets or decide whether a project passes. Instead, it supports the review process by comparing past project documents with new project plans to find projects with potential similarity or overlap and by drafting minutes or review opinion templates. It explains complex technical terms, summarizes lengthy project descriptions to highlight key points, and offers collaboration features so standing members can jointly draft and share project-by-project review opinions.
In the demonstration that day, the process of checking similarity with past ministry projects for a specific new project was presented. The AI provided a list of highly similar projects and summarized each project's purpose, content, budget, duration, and similarities and differences with existing projects. Based on this, the standing member can review the new project's distinctiveness by checking past budget requests, annual input amounts, and research results.
The Ministry of Science and ICT said it used budget requests, planning reports, and standing members' review opinions for about 5,000 national R&D projects accumulated over the past five years to train the AI. It also linked the system to about 12.43 million research outcome records from the National Science & Technology Information Service (NTIS). NTIS is a national databases that aggregates science and technology information such as government R&D tasks, research outcomes, papers, and patents.
In fact, standing members who used the specialized AI over the past three days said it allows much more discussion among members by reducing the time needed to draft templates, and that it helps in writing alignment with higher-level strategies, distinctiveness, connectivity, and overall opinions.
There were areas for improvement as well. The standing members suggested it would be more useful if features to check the logical consistency of drafts and interfaces to exchange views with standing committees in other technology fields were added. They said it is not enough for AI to write quickly; it should advance in a direction that consolidates judgments across fields and enhances the logical coherence of review opinions.
Kim Mimi, director of the Machine Information and Communications Coordination Division at the Ministry of Science and ICT, said, We will rapidly advance the system after piloting it this year, adding, At the final stage, the goal is to generate review opinions at the level of review officials. We plan to complete major advancement steps by next year.
The Ministry of Science and ICT is also reviewing options to expand the system so that each ministry can use it at the project planning stage for R&D projects. The aim is to check whether another ministry is preparing a similar project before launching a new one and, if needed, to develop it into a collaborative project.
Bae Kyung-hoon, Deputy Prime Minister and Minister of the Ministry of Science and ICT, said, Starting with this year's pilot, we will advance review support functions such as analysis of similar and overlapping projects and, going forward, develop the system so each ministry can use it in the process of planning R&D projects and requesting budgets.