ESTsoft wins JDC contract for in-house sLLM on internal servers
The six-month project covers 30B-class and 120B-class models, RAG search over 3.8TB of documents and a closed-network LLMOps setup.
ESTsoft said on 21 October 2026 that it has won the contract to build an in-house generative AI small language model (sLLM) environment for the Jeju Free International City Development Center (JDC), according to techm.kr.
The project is intended to raise employees' work accuracy and processing speed, and runs for six months from the contract date. ESTsoft will build the sLLM directly on JDC's internal servers, kept separate from the external network.
Under the plan, language models of 30B-class or larger will handle routine Q&A, while 120B-class or larger models will support complex long-form reasoning in specialized work-support chatbots.
Around 3.8TB of internal documents will be embedded for a RAG-based AI document search service linked to existing systems such as the ECM. ESTsoft will also supply an execution agent environment based on API calls and an Agent Builder for designing and running multi-agent workflows.
Security and monitoring systems, along with LLMOps tailored to the closed-network environment, will be established as part of the build.
ESTsoft's CEO, Jung Sang-won, said the project will demonstrate safe, practical use of generative AI in security-sensitive public institutions. The CEO has also pointed to the company's Alan service as the base for its agentic AI capabilities.
Quick answers
How long is the ESTsoft-JDC project?
Six months from the contract date, following the contract award announced on 21 October 2026.
What models will be used?
30B-class or larger models for routine Q&A and 120B-class or larger models for complex long-form reasoning in specialized chatbots.
Where will the AI run?
On JDC's internal servers, separated from the external network.