en.Wedoany.com Reported - KT has independently developed KT RAG based on Retrieval-Augmented Generation (RAG) technology. This technology aims to enhance the accuracy of AI search engines, with its differentiated advantages lying in its ability to be optimized for different industries and its virtuous data cycle structure.

At the "KT AI Agent Learning Session" held at KT Gwanghwamun Building WEST in Jongno-gu, Seoul on the 17th, KT introduced its Agentic AI advancement strategy and the core competitiveness of KT RAG. RAG is a technology that generates answers by retrieving and providing relevant documents, primarily used to prevent hallucinations in generative AI. KT leverages this technology to maximize search accuracy and AI task execution capabilities.

By developing a specialized RAG model in-house, KT has enhanced its technological competitiveness. Specifically, the Embedding Model is used to find documents most similar to user queries from vast amounts of data, while the Re-ranker Model determines the display priority of retrieved documents. KT has improved the performance of these two models to be better suited for Korean language search.

KT's developed modular-structured search engine can flexibly respond to the needs of various industries. For example, communication services involving lines and equipment require Graph RAG technology for correlation analysis, while the manufacturing sector needs Multimodal RAG technology capable of viewing and analyzing equipment drawings and photos. The key lies in combining various technical architectures to achieve industry-customized applications.

Another unique feature of KT RAG is its virtuous data cycle structure. By collecting actual service operation logs, analyzing search failure causes and user feedback, making improvements, and redeploying, a closed loop is formed. This system has been applied to the company's internal AI agent service, currently used by approximately 14,000 KT employees with a usage rate of 97%.

KT plans to provide high-completion AI services to actual industries based on its internal environment operation experience. Kim Junseok, Head of the Agentic AI Lab at KT AX Future Technology Institute, stated that to reduce hallucinations, accurate data for judgment is delivered through RAG technology, and AI models are additionally trained on specific domain information such as medical, financial, and legal fields. Furthermore, various attempts are made to reduce hallucination phenomena, such as incorporating industry experience into prompts or using "Harness Engineering" to strictly control AI.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com