[Big Trend] Startups developing their own LLMs in specialized areas such as law and healthcare
[More information about the companies featured in this article is available on Unicorn Factory’s big data platform, ‘Datalab’.]
/Image generated by Microsoft ‘Designer’
As the government moves to develop a national flagship large-scale AI model and promotes the ‘Independent AI Foundation Model’ project to secure technological sovereignty in AI (artificial intelligence), some startups that did not participate in the project are nevertheless making their mark in the LLM (large language model) sector with their own technological capabilities.
The startups belonging to the five teams named on the final list for the Independent AI Foundation Model project announced by the Ministry of Science and ICT on the 4th include a total of 18 companies, including △Upstage △Twelve Labs △Rebellion △Liner △FuriosaAI and △Wrtn Technologies. Although they suffered the bitter disappointment of being eliminated in this evaluation, startups such as △Mathpresso △Tomorrow Robotics △Pebblous and △Junction Med also took on the project through consortiums, expressing confidence in their AI capabilities.
There are also ‘hidden champions’ that did not apply for this project but are building differentiated technological capabilities by developing their own LLMs. Representative examples include △BHSN (law) △Every AI Korea (healthcare) △Miso (home services) △CLEVI (generative AI services) and △42Maru (lightweight LLMs).
They’ve cracked the difficult fields of ‘law and healthcare’
/Graphic by Im Jong-cheol
Legal tech startup BHSN recently launched the law-specialized LLM ‘Alibi Astro’. The law-specialized LLM ‘Alibi Astro’, recently launched by legal tech startup BHSN, stands out. Alibi Astro enhanced its expertise through CPT (continued pre-training) based on extensive legislation, case law and policy data. It was trained in the language structures and contexts used in actual legal practice through RLHF (reinforcement learning from human feedback), incorporating the opinions of experts such as lawyers.
It can understand the context of questions, infer logical relationships between documents, and enable AI-based contract work, while reviewing even English construction contracts (EPC) averaging 100 pages in just one minute. It can build expert-level reasoning, including interpreting clauses and suggesting directions for revisions. Major Korean conglomerates, including CJ CheilJedang and Aekyung Chemical, currently use Alibi Astro in their daily operations.
e1-M, EveryAI Korea’s LLM specialised in the medical field, is also attracting attention. Unlike existing global models that rely on English-based translation, e1 is designed to intuitively understand and reason through the context and nuances of Korean.
It scored 90.78 on the medical QA benchmark ‘KorMedMCQA’ developed by a KAIST research team, outperforming GPT-4o (85.61) and Claude 3.5 (86.51). It was also highly rated for its reasoning ability to comprehensively analyse patient symptoms and test results and even suggest treatment directions.
EveryAI Korea plans to develop industry-specific LLMs for fields including law (e1-L) and finance (e1-F), in addition to medicine, and introduce e1-M-based medical AI solutions for use by hospitals, pharmaceutical companies and insurers.
“Why use ChatGPT?”… Low-cost, high-efficiency LLMs
There are also startups challenging ChatGPT with low-cost, high-efficiency LLMs. The generative AI service ‘Ivy’, launched by CLEVI, consists of small models created to minimise AI development costs, which are then distilled and combined. It introduced technology for efficiently managing the development process by applying automated, container-based deployment technology for training nodes (computing resources). It is characterised by significantly reducing service development and operating costs through efficient training pipeline design and optimisation of the communication network between nodes.
Miso, a startup providing more than 200 home services, including cleaning, moving and appliance rentals, is also transforming into a technology company by launching the industry’s first LLM-based solution in which AI suggests the home services customers need and supports connections with professionals. Whereas customers previously had to search for and apply for home-service categories one by one, the process has now evolved so that customers can discuss the problems they are experiencing, after which AI analyses them and recommends and connects them with the optimal service.
FortyTwoMaru achieved technological differentiation with a highly reliable LLM. Its proprietary ‘LLM42’ is a lightweight LLM optimized for the Korean language environment, characterized by its low-cost structure, high security and broad applicability in industrial settings.
Experts agree that the government should develop a strategy that positions not only companies participating in the independent AI foundation model project, but also other startups, as central pillars of the national AI strategy. Choi Seong-jin, CEO of the Startup Growth Research Institute, emphasized, “Institutional arrangements must be designed so that startups can stand as central players, rather than mere supporting actors, in AI policy,” adding, “The success or failure of an AI superpower depends on startups.”
/Graphic=Yoon Seonjeong
Reporter Choi Tae-beom [email protected]
Source: MoneyToday (Reporter Choi Tae-beom) | https://www.mt.co.kr/future/2025/08/07/2025080518330143511
[MoneyToday startup media platform ‘Unicorn Factory’]
<Copyright © MoneyToday, ‘Real-time news where money is visible’. Unauthorized reproduction, redistribution and use for AI training prohibited>