NEWS

Getting AI to Use Company Data Right: Monthly AX October Seminar Draws a Full House

View from behind the audience at the Monthly AX October seminar. The screen shows the title slide of Genesislab's talk on lightweight ontology and AI governance.

Source: IT Chosun (IT조선) — Genesislab: "Build and Revise AI Work Knowledge Quickly with a 'Lightweight Ontology'" (Monthly AX, October Issue) Original in Korean

Source: IT Chosun (IT조선) — "Is Connecting Data to AI Enough? Understanding the Work and Controlling Execution Are What Matter" (Monthly AX, October Issue) Original in Korean

The October edition of the Monthly AX seminar took place on November 3 at Leaders Hall, in the Hanbit & headquarters in Seodaemun-gu, Seoul. The theme was "Our Company's Data: How Can AI Use It Correctly and Safely?" Every seat was taken before the first talk began, a sign of how many corporate practitioners are working through data connections and AI governance.

Co-hosted by IT Chosun and Hanbit &, the seminar featured talks from three companies in turn: Oracle Korea, Genesislab, and Arize. All three started from the same concern, which is that attaching data to AI does not produce business results on its own. Genesislab built Slipbox to solve this problem. Slipbox is a knowledge management system that helps AI understand what internal data means and how it connects. It had stayed inside the company until this seminar, where Genesislab showed it publicly for the first time.

Genesislab CTO Daehun Yoo presenting at a podium, with a poster for the October Monthly AX seminar in the background.

Genesislab CTO Daehun Yoo presenting at the October Monthly AX seminar

Why AI Gets Wrong Answers from Internal Data

Daehun Yoo, CTO of Genesislab, spoke about lightweight ontology and AI governance. Corporate databases are full of abbreviations and code values that only the people in charge understand, and the relationships between data are complex. Even as LLMs improve, an AI that misreads this content can run a query without any error and still return the wrong answer.

An ontology organizes the meaning and relationships of data so that AI can understand them. But an ontology is never finished. It needs continual adjustment as agents are built and used, and when its structure is complex, people have trouble finding what to change.

Managing Knowledge in Documents People Can Read and Edit

To ease that burden, Genesislab added Slipbox to Agentria, its AI agent platform. Slipbox takes a lightweight ontology approach. Instead of a complex graph, it keeps business terms, data relationships, and rules in documents people can read and edit. The documents link to one another, so AI can find and use the knowledge it needs.

No one has to write these documents from scratch. Slipbox analyzes the structure of existing databases and past query history to generate knowledge pages automatically, and the people in charge review and correct them. Companies do not need to replace their systems wholesale.

In an internal demo, the 121 tables of a fictional manufacturer's ERP produced 135 knowledge pages and 223 pieces of relationship information. These results were measured on a development server.

Genesislab also shared an accuracy comparison. Oracle's natural language query feature, Select AI, scored 56% accuracy on its own and 100% once Slipbox was connected. This was a small internal comparison of eight questions, each tested twice, so it does not mean the same result would hold across real work. A query about training products shows the difference clearly. With Select AI alone, the answer was that no such data existed. With Slipbox connected, the system found Training, the code value actually stored in the database, and retrieved 3,901 records.

Knowing the Rules Does Not Mean AI Will Follow Them

Yoo said that knowing the business does not make AI safe to trust with the work. Tell an AI in plain words to get approval, and sometimes it will comply and sometimes it will not.

Genesislab pairs Slipbox with Ability. The AI chooses the task it needs, and Ability runs that task according to a predefined procedure. For work where the approver depends on the contract amount, the approval chain and procedure are enforced in code rather than left to the AI's judgment alone.

Yoo also listed what real-world operation requires. Companies have to manage which agent did what with which permissions, track and audit those actions, and be able to roll back mistakes.

Agentria Trust Handles Legal and Certification Work

Genesislab also introduced Agentria Trust, its AI governance product. It manages supporting evidence against AI laws and certification standards and helps AI check in advance whether control items have been met. The product grew out of Genesislab's own experience. When the company prepared for an AI trustworthiness certification in 2024, gathering materials alone took about a month, and the on-site review took five days.

"Agents should be given business context, and people should be given clear control," Yoo said. In this plan, Slipbox manages the business knowledge given to AI, Ability controls execution procedures, and Trust supports legal and certification work.

What Oracle and Arize Covered

The three talks moved from data architecture to ontology and governance, then to agent monitoring.

Minji Kim of Oracle Korea presenting at a podium

Minji Kim, Cloud Specialist Engineering Team, Oracle Korea (Photo: IT Chosun)

Minji Kim of Oracle Korea explained that for AI to judge accurately, collecting data is not enough. It also has to know how pieces of information relate and what the company's rules are. Kim also introduced Actionable Ontology, a concept in which AI assesses a situation and goes on to suggest follow-up actions.

Seungmin Lee of Arize presenting at a podium

Seungmin Lee, Head of Asia Pacific and Japan (APJ) Solutions, Arize (Photo: IT Chosun)

Seungmin Lee of Arize covered the security risks that come with agents taking on real work. Spelling out prohibitions is not enough, so access has to be limited in advance and agent actions have to be logged. Arize presented a four-stage framework of collecting execution records, assessing risk, alerting on anomalies, and taking follow-up action.

The talks were followed by Q&A and open networking. More details are in the two IT Chosun articles linked above.

List

1-3 Page Myeongdong, 5th Floor, Myeongdong 1-ga, Jung-gu, Seoul Metropolitan City | CEO Lee Young-bok | Business registration number 421-88-00471 | Mail-order sales registration number 2017-Seoul Jung-gu-1784 [Check business information]
Contact number: 02-6402-0118 (Operating hours: Weekdays 11:00~18:00) | Email contact@genesislab.ai | Hosting provider Genesislab

© 2026 Genesislab, Inc. /

/

5F, Page Myeongdong, 1-3 Myeongdong 1-ga, Jung-gu, Seoul | CEO Lee Young-bok | Business Registration Number 421-88-00471 | Mail-order Business Report Number 2017-Seoul Jung-gu-1784

Inquiry phone: 02-6402-0118 (Business hours: Weekdays 11:00~18:00) | Email Sales@genesislab.ai | Hosting provider Genesislab

© 2026 Genesislab, Inc. /

/

1-3 Page Myeongdong, 5th Floor, Myeongdong 1-ga, Jung-gu, Seoul Metropolitan City | CEO Lee Young-bok | Business registration number 421-88-00471 | Mail-order sales registration number 2017-Seoul Jung-gu-1784 [Check business information]
Contact number: 02-6402-0118 (Operating hours: Weekdays 11:00~18:00) | Email contact@genesislab.ai | Hosting provider Genesislab

© 2026 Genesislab, Inc. /

/