Lee Yoon-pyo(Department of Nuclear Engineering, Senior) collaborated with the Korea Atomic Energy Research Institute(KAERI) as an intern in the goal 토토사이트 Research Lab to jointly develop an goal 토토사이트 agent assistant for reactor operation. Lee took charge of core development tasks, including overall workflow design and operational logic development.

The newly developed goal 토토사이트 agent, built on the ChatGPT 4.0 API and Python, utilizes a large language model(LLM) to enable assistance for natural language-based reactor operation. Whereas existing goal 토토사이트 systems merely execute pre-learned procedures, this system can respond dynamically to user commands in natural language.

Its functionality was validated through simulation. The AI agent successfully handled scenarios of boron concentration anomaly within the iPWR environment—an SMR(Small Modular Reactor) simulator provided by the International Atomic Energy Agency(IAEA). Also, it was shown capable of generating and executing new operational procedures upon command. “Currently, the system utilizes the external API, GPT-4, but it will be integrated with KAERI’s proprietary ‘Atomic GPT’ under development for further enhancement,” Lee explained.

Lee attributed the success of this project to his foundational knowledge in thermohydraulics and reactor engineering from his studies of his major, nuclear engineering, in combination with self-taught AI skills. “I wanted to explore the intersection of artificial intelligence and my major,” he said, explaining his motivation for applying. Though he had no prior experience in developing AI, he taught himself PyTorch before applying and further built his capabilities during the internship by studying research papers and open-source materials.

He emphasized the significance of him participating in research as an undergraduate. “Though the initial phase of learning AI development and programming was especially challenging, the experience at KAERI has been an invaluable stepping stone to me as a researcher,” he reflected.

Lee is now preparing to pursue a doctoral course in nuclear AI in the United States. He mentioned his intention to focus on developing autonomous operator support systems and explainable AI. “Currently, human operators are responsible for reactor operation,” he said. “Rather than replacing them, AI should evolve to assist and complement their work.” He added that, “Since operator labor costs are not a major portion of reactor operation expenses, AI should aim to improve operators’ convenience rather than replace them.”

He also emphasized the importance of interpretability in safety-critical fields such as nuclear energy. “It is essential to understand why AI arrives at a specific conclusion,” he noted, describing his prior work using the GEMMA model to train nuclear-specialized AI and analyze its reasoning process.

This research was published in the international journal Nuclear Engineering and Technology under the title, Large Language Model Agent for Nuclear Reactor Operation Assistance, with Lee listed as the first author.