RUNICON

Japanese workflows as AI training data.

We’re preparing worked examples and environments where AI models can practice the same tasks.

Join the launch waitlistExplore the cases

The challenge for LLM teams

Imagine opening a workbook and finding the answer pages missing. Teams training AI on Japanese professional work face a similar gap.

WorkbookThe questions are there
Worked answersThe solutions are missing
Source documents may be available. What’s hard to find is the expert’s finished work together with an explanation of the choices behind it.

Train your model on the work you choose

  1. Request a task

    Tell a specialist which job you want your model to learn.

  2. Get a worked answer

    Train on the source material, a finished example and the reasoning behind it.

  3. Let your model practice

    Run the task again and see where your model loses points.

The task collection

A banker reviewing financial statements

Update a sales spreadsheet

The task
Read a company’s earnings report and enter this year’s sales in its spreadsheet.
Example answer
Enter ¥12.4 billion as 124 in a column measured in units of ¥100 million.
View the document specimen
Illustrative document
Earnings release · extract
12,400 JPY million
Workbook unitJPY 100 million
Model input124.0
Example dataset for purchase

An earnings report, a blank sales worksheet, the completed sheet and notes tracing each figure and unit conversion.

Next case
Construction professionals checking drawings on site

Revise an estimate when a drawing changes

The task
Compare the old and new drawings, then update the quantities and costs.
Example answer
When a wall grows from 18 to 24 m², add the cost of the extra 6 m².
View the document specimen
Illustrative document
Drawing A-12 · partition wall
Revision B → C18 m² → 24 m²
Quantity change+6 m²
Example dataset for purchase

Original and revised drawings, both estimates and a breakdown of the quantity and cost changes.

Next case
Physicians reviewing handover notes on a hospital ward

Write a handover for the next hospital

The task
Read a fictional patient’s notes and list what the receiving team needs to follow up.
Example answer
For a pending test, write “Result pending. Receiving physician to follow up.”
View the document specimen
Illustrative document
Transfer note · fictional case

Specimen collected. Result pending at transfer.

Follow-up assigned to the receiving physician.
Example dataset for purchase

Fictional patient notes, a completed handover and an explanation of which details belong in it and why.

Explore the planned offering

What you can get at launch

Buy a training dataset
Get tasks, source files, worked answers and explanations in the field you need. Use them to train your own model.
Use an RL environment
Have your model attempt tasks and use its scores as rewards for repeated practice.
Commission a custom task
Ask a specialist to write a problem, reference answer and scoring criteria for a workflow you specify.