Generative artificial intelligence (AI) is evolving beyond simply answering civil complaints to designing and guiding users through complex administrative procedures. A research team at Yonsei University's Data Science Lab has proposed a platform in which multiple AI agents collaborate to handle complex civil requests, presenting new possibilities for AI-based public administrative services.
Yonsei University said Monday that a student team from its Data Science Lab—Shin Dong-jun, Kim Seo-yoon, Kim Jung-min, Kim Si-won, and Lee Ji-won, under the supervision of Professor Park Tae-young—won an Excellence Award at a competition hosted by the Ministry of the Interior and Safety for AI-based civil service innovation scenarios and development methods.
The competition was organized to discover ways to innovate civil services that citizens can tangibly experience using generative AI. The Yonsei team earned high marks for proposing 'IRMA (Integrated RAG Multi-Agent),' a multi-agent civil service platform based on integrated retrieval-augmented generation (RAG).
IRMA is characterized not by a single AI model handling all tasks, but by multiple specialized AI agents dividing roles and collaborating. A general agent that analyzes the user's intent works alongside a knowledge agent that searches and verifies relevant laws and administrative information, and an execution agent that analyzes permit and licensing procedures to present checklists and action plans. Together, they provide one-stop support for complex civil requests involving multiple agencies.
In particular, IRMA is designed to go beyond simple question-and-answer, comprehensively assessing permit procedures and eligibility requirements according to the user's situation, and providing the necessary administrative steps as a customized action path in timeline form. It also implements security and management features required by public institutions, including a State Store that maintains consultation context, centralized log management, and automatic personal information masking (PII Masker), enhancing its potential for real-world administrative service applications.
"AI must go beyond simply generating answers to support the entire process through which citizens actually use administrative services," said Park Tae-young, the supervising professor at Yonsei University's Data Science Lab. "This research is significant in that it presents the potential of AI administrative services that help anyone easily understand and execute complex civil procedures."
The Yonsei Data Science Lab team that took part in the competition said, "Existing civil services placed a heavy burden on citizens, who had to find information and judge procedures themselves. Through IRMA, we sought to propose a new AI-based civil service that supports this burden at the system level with AI, so that anyone can easily understand and execute complex administrative procedures."
