Breaking into regulated industries with legal tech: "We will build a 'legal assistant for everyone'"

[For more diverse corporate information on startups, you can view it on the Unicorn Factory big data platform's 'Data Lab'.]

In the fields of law and finance, AI's "plausible wrong answers" can lead to accidents. In the United States, lawyers have repeatedly faced sanctions for submitting fabricated case precedents generated by AI to courts. It is a domain where missing even a single clause or definition provision can reverse the conclusion. This is why it is difficult to immediately apply general-purpose AI to such regulated industries, no matter how intelligent it becomes.

Mentat has exploited this gap. Kim Ju-su (CEO), a founding member of financial AI company Earnest AI, said, "We are a company that supplies domain knowledge unknown to general-purpose AI—namely 'what' and the way experts work, 'how'—to AI," adding, "We build 'expert AI' for regulated industries where errors are not permitted."

The starting point of Mentat's technology is data. Law firms and financial institutions have accumulated documents in various formats, including PDFs, Hangul files, Excel spreadsheets, and audio files. Humans can read them, but AI cannot process them properly. Mentat converts these into "data that AI can read" using its own converter specialized for regulated industries. It has also layered on a search algorithm created by pre-analyzing the types of questions that may be asked. Kim (CEO) explained, "If the input data is poor, AI's answer will inevitably be poor," adding, "We have combined creating good data with effectively retrieving and using it."

The legal knowledge graph 'Codex' is also a core technology. Laws involve chains of delegation through enforcement decrees, enforcement rules, supervisory regulations, and appendices. General-purpose AI often stops at the law or enforcement decree and may skip the definition provisions at the beginning of the law to interpret it arbitrarily. Codex connects these links seamlessly and retrieves them all at once. Answers include the original text from the National Law Information Center, enhancing the convenience of "fact-checking."

Last month, the company officially launched 'Mentat AI,' a document drafting service for lawyers. By uploading case materials and inputting strategic directions, it drafts initial versions of complaints, responses, and preparatory documents. Kim (CEO) stated, "We take pride in saying that our document drafting quality is the best among general-purpose AI and domestic legal tech companies."

The first major client was Law Firm Taipyongyang. The company is developing an AI agent that converts the vast legal materials held by Taipyongyang into data for AI use and finds past similar cases and related documents. Kim (CEO) said, "Previously, it required considerable time and effort to find necessary information from massive amounts of materials," adding, "Our goal is to reduce the time spent on document search and basic review using AI, supporting lawyers so they can focus on more advanced legal analysis and strategic judgment."

The customer base is also expanding into the financial sector. This year, Mentat was selected for 'Dino Lab Tech 1st Cohort,' an open innovation program solving Woori Financial's technical challenges, and is creating a customized tax-saving guide (AI Asset Care) based on tax law information. It is an agent that simulates tax-saving strategies for tax-benefited products such as pension savings, IRP, and ISA. With major property and casualty insurance companies, it is testing an agent that links terms and conditions, internal guidelines, case precedents, and dispute mediation decisions into a single knowledge graph to respond quickly to complaints. It is also discussing plans with savings banks and capital companies to automate compliance tasks such as advertising review.

In its first year of business this year, revenue has already exceeded 2 billion won and is expected to close around 3 billion won. Mentat AI gained 300 lawyer sign-ups within one month of launch. The goal is to reach 5,000 by the first half of next year.

Kim (CEO) said, "Our goal is to become a 'legal OS (operating system)' that encompasses various fields." The plan is to take on the entire legal and compliance work of each industry, just as litigation lawyers assist with lawsuits. In finance, the goal is to select one sector and handle its entire compliance. Insurance is the most promising field. Kim (CEO) explained, "Insurance inherently has a high potential for disputes because tens of millions of won can hinge on whether payment is made, making back-end compliance tasks heavy." It is also discussing application plans with partners in the construction and medical fields, which face criminal penalty risks under laws such as the Serious Accidents Punishment Act.

Codex includes everything from past laws to bills in legislative notice, targeting the needs of large law firms or corporations that must read the direction before a law is finalized. A dedicated organization led by PhD-level researchers is separately enhancing Codex. Kim (CEO) noted, "Even if all are publicly available original legal texts, comparing pre- and post-amendment versions chronologically and maintaining the links requires considerable technology."

The company is also opening its doors to places without lawyers. It plans to supply a legal research service with the document drafting function removed to companies without in-house counsel or media outlets.

Mentat refers to humans with computer-like computational abilities in the SF novel 'Dune.' Kim (CEO) said, "I want to build an agent like a true 'Mentat'—a soul partner that understands my work best and executes it accurately even if I speak vaguely," emphasizing, "We will also increase legal accessibility for people who have given up on litigation due to costs."

[MoneyToday startup media platform 'Unicorn Factory']