Your team is the hero. Our agents work beneath.

We design, build and run AI agents that read your documents, answer first-line questions and search your files.

What we build

Six kinds of agent work, plus a two-week sprint to choose where to start.

Recent work

All work

A casting platform in Japan2025

Talent-profile PDFs, straight into the database

A pipeline that takes actor-profile PDFs in any layout by email, splits them per person, extracts about 20 fields and registers each profile on the platform.

Outcome
94.7%
field accuracy on the regression test (an 11-profile PDF)

A tire retailer in Japan2025

A sales assistant that helps customers choose tires

A chat assistant that asks about the car, tire size and priorities, recommends two or three products from the store’s own catalog, mentions current campaigns and records the lead.

Outcome
2–3
recommendations per conversation, all from the store’s own catalog

Taiyaki Studios (our founder is on the team)2024–2026

AI pipelines for translation and animation

A translation tool with human review, and a pipeline that turns webtoons into animated episodes, for a comics app.

Outcome
~90%
of AI translations accepted as-is in human review

How a project runs

  1. Talk

    Tell us how the work runs today and where it hurts, in a free 30-minute call. If AI is the wrong tool, we say so.

  2. Discovery sprint

    One to two weeks building a prototype on your own data. We measure accuracy and time saved, then quote a fixed price for the build.

  3. Build

    We connect it to the tools you already use, add permissions, logs and a review screen for your staff, and ship it.

  4. Run and improve

    After launch we keep measuring accuracy and fix what breaks. Your team keeps the final check and the decisions.

縁の下の力持ち

縁の下の力持ち (en-no-shita no chikaramochi) is what Japanese calls the person who holds a house up from under the floor. That’s where our agents work: they read the documents, find the files and draft the replies, then hand them up to your team.

FAQ

How do you handle our data?

We use your data only for your project. We choose AI services whose terms and settings exclude training on your data, and we tell you which models we’ll use before we start.

Which AI models do you use?

We choose between OpenAI, Google Gemini, Anthropic Claude and others based on the task, accuracy, cost and data requirements. We’re not tied to any vendor.

How much does it cost?

It depends a lot on scope, so the discovery sprint ends with a fixed quote. We’ll tell you what the sprint itself costs in the first call.

Can we work in Japanese or English?

Yes, either. The systems we build are tuned for Japanese business documents, from era dates to honorifics.

Who owns what you build?

You receive the full source code and design documents. Ownership is set out clearly in the contract.

Do you work with small companies?

Yes. Starting with one small workflow is often the best way in.