AI and generative AI product development
Beyond the proof of concept,
into everyday use.
We build generative AI into products and into day-to-day work, and we take it from planning through design, implementation and operation. It is designed from the start to reach the point where people actually use it, not to end at a demo. We run AI in our own development and our own back office, so we can judge from experience where it can be trusted and where a person still has to look.
Where our clients usually are when they call
- 「We want to use AI but can't tell what it can and can't do」
- 「We built a proof of concept and it never made it into the business」
- 「Answering enquiries and drafting documents eats our time」
- 「Our documents are scattered and nobody can find anything」
- 「We want AI in our existing system but don't know where to cut in」
What we do
AI agents and workflow automation
We hand the routine steps to AI and leave your people the judgement calls. It's wired into Google Apps Script or the workflows you already run, so it lives inside the working day.
RAG over your own documents
Search across scattered internal material and get answers with the source attached. Because it shows what it drew on, an answer can be checked rather than trusted blindly.
Chatbots
For internal questions or first-line customer contact. We design the hand-off to a person for anything it can't answer, rather than leaving it to guess.
Adding AI to systems you already run
Summarising, classifying, drafting, added to something already in production. We work out first how much can be done without a rebuild.
Building generative AI products
Services with AI at the centre, from concept through to production. We can prove it out first and move to production only once it holds up.
Setting your internal ground rules
Which AI is allowed, and what must never be typed into it. Deciding this before the tools arrive keeps things from stalling later.
How it goes
- 1
Free consultation
You tell us what you want to achieve and what's difficult now. It's fine if none of it is organised. If AI is the wrong fit, we say so up front.
- 2
Build an MVP
We don't build it all at once. We build the smallest version that proves the effect. If it doesn't hold up here, not proceeding is a real option.
- 3
Build for production
We build it into something usable, informed by what the trial showed. Where accuracy isn't enough, a human checkpoint stays in the flow so operations don't stall.
- 4
Run it and improve it
We adjust as we see how it's used. Models and pricing change, so switching is part of the conversation too.
How we choose the technology
Both the model and the cloud are chosen per project. Different models are good at different things, so we compare accuracy, speed, cost and how your data is handled before recommending anything. Reading long documents, returning strictly structured output, and processing at volume all point to different choices. If you already run something, we work with it; if you want Claude, or a particular platform, we design around that. We often mix models rather than commit to one. We are a Google Cloud Partner, but that is never a reason for us to steer you to Google's products. You can also name something that isn't on this list.
Questions we get
- We don't know what to ask for. Where do we start?
- Start with "this task is eating our time". From there we separate what AI can take off you from what it can't. We're happy to talk before you have a goal defined.
- Is it safe to put our own data into AI?
- Whether your input is used for training depends on the plan you're on. Japan's Personal Information Protection Commission asks that, where personal data such as a customer list is entered without the individual's consent, you satisfy yourself that the provider will not use it for machine learning. We work out with you what may and may not go in.
- Can we start with an MVP?
- Yes. Building the smallest version that proves the effect, then moving to production only if it holds up, is our default. We quote on the assumption that not proceeding is a legitimate outcome.
- Can you run it after it's built?
- Yes. With generative AI both the models and the pricing move, so switching and tuning are part of the work. Operations can be included.
- How much does it cost?
- It depends on the scale and the scope you hand over, so we don't publish a single figure. We agree the scope at the first consultation and quote from there. You are welcome to get competing quotes.
- Where do you work?
- We work online with clients across Japan. If you are in or near Tokyo we can come to you. We work in Japanese, English, Korean, and Spanish.
Let's find out what AI could take off your plate.
The first consultation is free. It's fine if the goal isn't defined yet. We usually reply within 1 to 3 business days.
Book a free consultation