AI service specialist company ESTsoft (CEO Kim jang jung) announced on the 17th that it has officially released the large-scale language model (LLM) 'Allen LLM' developed in-house, completed the commercialization of the LLM product, and began providing full-scale services.
Allen LLM is an open-source-based model specialized in report generation and inference based on search augmentation generation (RAG), reflecting the know-how of the currently serviced AI search engine Allen. To this end, it demonstrates powerful performance in advanced fact-checking, search result filtering and ranking optimization, agent execution plans and tool calls, as well as understanding Korean context.
With the launch of this in-house model, ESTsoft has established itself as an LLM company that simultaneously realizes technology internalization and practical commercialization. In particular, by developing an in-house LLM optimized for AI search engines, an independent technical ecosystem for Agentic AI service has been built, aiming to challenge the government-led ultra-large AI development project 'World Best LLM (WBL)'.
In terms of ESTsoft's LLM commercialization and WBL participation, the Allen LLM advanced environment is expected to act very favorably. This is because Allen LLM has already been embedded in its AI search engine service (alan.est.ai), and expansion to external projects is planned. Through this, the plan is to rapidly elevate the performance and efficiency of Allen LLM to the level of commercialized global LLM by providing specifications tailored to various service environments.
ESTsoft has configured Allen LLM from ultra-large models for data centers to lightweight models for on-device usage. It has been designed from the start to be adaptable flexibly to various business environments and service scales.
First, the ultra-large model provides over 200 billion parameters for environments requiring large resources, such as building search-based agent services. The Allen LLM's ultra-large model enables rapid and stable AI agent construction in specific domains or closed networks and intranets, meeting the data protection requirements demanded by many companies and institutions.
The lightweight model is utilized for situations requiring quick responses, such as customer service, or for expansion to on-device usage. In fact, the lightweight model is already being used in the company's AI live chat service and AI kiosks. In the future, the plan is to expand its application to various devices.
Allen LLM plans to gradually make the lightweight model open source in the second half of the year.
Kim jang jung, CEO of ESTsoft, stated, “The launch of Allen LLM signifies that ESTsoft is advancing its AI business centered around language models, beyond vision AI,” and added, “We plan to accelerate our customized solution business in various industries and service environments, based on the Agentic AI and LLM technology of proven AI search engine Allen in B2C.”
Meanwhile, ESTsoft has consistently demonstrated its technical prowess in the AI field through various challenges. It recorded first place domestically and third place globally on the ‘Open Ko-LLM leaderboard’ operated by the National Intelligence Information Society Agency to evaluate Korean model performance, and won the final victory in Season 1 of the AI Grand Challenge hosted by the Ministry of Science and ICT for three consecutive years, receiving the Minister's Award.