KT has successfully completed a pioneering project with Jasmine Technology Solutions (JTS) of Thailand's Jasmine Group, establishing a comprehensive Thai-language Large Language Model (LLM) platform. This "one-stop project," spanning over a year since its inception in March 2023, encompasses three core components: the joint development of a specialized Thai-language LLM, the implementation of an LLM operation management environment dubbed 'LLM Ops', and the establishment of a dedicated AI service infrastructure known as 'GPU Farm'.

The Thai-language LLM was co-developed by KT, leveraging its proprietary model development experience, and Upstage, a Korean LLM specialist in which KT had invested in 2023. This model is designed to support Thai, English, and Korean languages. Its core capability stems from extensive training on a vast corpus of proprietary Thai data, enabling it to accurately respond to complex queries related to Thai politics, history, and culture. This suggests a foundational pre-training phase on a large-scale, diversified Thai dataset, potentially followed by domain-specific fine-tuning or augmented generation capabilities leveraging the cultural nuances embedded in the local data.

The LLM Ops system represents a critical operational methodology. It is an end-to-end management environment built for the entire lifecycle of LLM models, encompassing training, deployment, and ongoing operation. Customized for JTS, LLM Ops provides a streamlined and efficient framework for managing the intricate processes inherent in generative AI environments. This system likely incorporates MLOps principles tailored for LLMs, including data versioning, automated model retraining pipelines, model serving infrastructure with API gateways, performance monitoring, and potentially A/B testing capabilities for model updates, ensuring efficient and stable operation.

Complementing the LLM and its management system, KT Cloud, KT's cloud subsidiary, established a GPU Farm in Thailand. This AI service infrastructure focuses on managing high-performance GPU resources. The technical underpinning of the GPU Farm involves setting up and operating data centers equipped with powerful GPUs, managed by a robust orchestration layer (e.g., Kubernetes, Slurm) to facilitate efficient resource allocation, job scheduling, and multi-tenancy. This infrastructure will enable JTS to offer GPU as a Service (GPUaaS) to Thai enterprises and institutions, fostering the expansion of the local AI ecosystem by providing scalable and accessible computational power.

This project marks KT Group as the first Korean enterprise to build such a comprehensive AI infrastructure overseas for the development, operation, and expansion of generative AI services. It also serves as a significant case study of successful global AX (AI Transformation) business collaboration between a large corporation and a startup. KT intends to continue its partnership with JTS, focusing on AI platform advancement, GPU infrastructure expansion, maintenance, and the exploration of new AI-driven services, aiming to progressively introduce AI services tailored for the Thai AX market and beyond into other global markets. Furthermore, KT plans to diversify its AI and cloud service offerings, building on this experience and leveraging strategic alliances with global tech leaders like Microsoft and Palantir.