Skip to content
Class NK × MetaWeave
Get the app
macOS Windows iOS

How AI can help in classification work — a Class NK lens

Notes and demonstrations from a conversation between MetaWeave and Class NK, hosted by NBS (NYK Group). Three patterns where AI orchestration could fit alongside classification work — and the working examples we showed in the meeting.
Three patterns where AI could help The context we bring GitHub
One person at a white desk surrounded by small kawaii agents, each working on a sub-task. The person directs; the agents execute.
A surveyor — or a rule editor, or a customer-service coordinator — supported by a small team of agents. The person decides; the agents do the volume.

Three patterns we showed

01

A knowledge layer

A natural-language search over a working corpus of 10,000+ pages of Class NK rules, plus SOLAS and MARPOL. Every answer cites the chapter and paragraph. A natural extension adds an IMO change-tracker chain on top.

02

Validation at scale

An agent team that handles the routine parts of validating client data submissions — sitting naturally alongside ZETA, with the surveyor at every judgment seat.

03

Department agent teams

The same orchestration pattern, applied per department. An agent team that watches, suggests, and drafts; the department’s people approve, edit, decide.

The context we bring

Before any of these patterns, we have been operating on Class NK data for several years — on the ship-operator side. We scrape the Class NK portal for ship-operator clients. We track Class NK certificate lifecycles. We coordinate around survey windows and follow defects to closure with Class NK surveyors.

That existing context is what informed the three patterns above. They are not abstract proposals — they are derived from watching Class NK’s processes from the outside, daily, for the operators we serve.

The six ship-side workflows we already run on Class NK data →

How AI has changed in 24 months

2023

Conversation

Single conversations. No memory. No actions.

2024

Action

Single agents could do single tasks. Often unreliable for enterprise work.

2025–26

Teams

Agent teams work together — handoffs, memory, audit. Reliable enough for production. This is now.

How we work — briefly

Source-available

Anything we build is owned as code. Forkable, modifiable, no vendor lock.

Internal hosting

If a capability is deployed, it runs on your VM. Your data stays inside your perimeter.

Working alongside NBS

NYK Business Systems is deploying the same substrate in Tokyo this week. That deployment is the working reference for how we work.

“In the future, we hope to establish a structure where NK hosts technical workshops, MetaWeave acts as a technical consultant, and companies like NYK/NBS participate as attendees.” — Mimura-san, NBS, 7 April 2026

If a conversation continues →
Three patterns where AI could help A knowledge layer · validation at scale · department agent teams. Three areas, with working examples.
The context we bring Six ship-side workflows we already run on Class NK data, for ship-operator clients.
Demo walkthrough The validation agent team end-to-end, with commentary.
If a conversation continues Notes on how a working conversation could continue — through Class NK's existing patterns.