Korean AI models expand into calling, translation and vehicle control
South Korean companies are deploying homegrown AI models across industrial applications, from customer service calls to autonomous vehicle features. The Ministry of Science and ICT said on May 11 that organisations ranging from LG AI Research and SK Telecom to startups like Upstage and Motif Technologies are accelerating AI transformation with domestic models.
Call centers and voice phishing detection
LG AI Research deployed its Exaone model in LG Uplus' AI agent IxiO, which handles incoming calls. The system transcribes conversations and analyses context to generate summaries with recommended follow-up actions.
IxiO organises information customers need after calls-promised discounts, agreed unit prices, gift items-without manual entry. The system also detects voice phishing in real time, flagging psychological pressure tactics and alerting users to risk.
Yu-cheol Kim, head of the strategy division at LG AI Research, said the goal is for "AI technology to quietly seep into people's everyday lives" rather than demand attention.
Real-time translation services
Flitto, an AI data and solutions company, integrated Upstage's Solar Open model into its Chat Translation service for real-time interpretation. Solar Open is an open-source model developed with government support.
Combining Solar Open with Flitto's multilingual datasets improved both translation accuracy and processing speed in live service environments.
In-vehicle voice control
SK Telecom released A.Dot Auto, an in-vehicle AI agent built on its A.Dot X model. Drivers can use voice commands to navigate, play music, control air conditioning and search for information without taking their hands off the wheel.
The system uses an agentic AI approach, processing multiple vehicle functions sequentially in response to single voice commands. SK Telecom plans to extend this approach to other industries beyond automobiles.
Math education and logical reasoning
Mathpresso is integrating Motif Technologies' large language model into QANDA, an AI math learning service used by millions globally. The service lets students photograph problems and receive step-by-step explanations.
Motif's model has 300 billion parameters and uses proprietary training methods to guide the system toward logical reasoning. The company first validates mathematical performance in smaller models before scaling to larger ones.
Sung-min Lee, head of the AI group at Motif Technologies, said the model "implements the path of deep thinking that humans go through when solving complex problems."
Central banking and financial analysis
The Bank of Korea deployed Voki, a generative AI platform built on Naver Cloud's model. It's the first such deployment at a central bank globally.
Voki searches and summarises economic and financial documents, answers questions about them, and supports analysis of economic issues. The Bank of Korea plans to fine-tune the system with its own data to create a finance-specialised model.
Naver Cloud said the deployment demonstrates that domestic AI models can meet security and trust requirements in the public finance sector.
Learn more: Generative AI and LLM courses cover foundation models like those described here. AI Agents & Automation training explores the agentic approaches used in IxiO and A.Dot Auto.
Your membership also unlocks:
