Questions we get, answered properly.

Send a few paragraphs describing the problem to the contact form or straight to our inbox. You will get a reply from an engineer, not a salesperson. If it looks like a fit, we book 30 minutes; if it does not, we say so and point you somewhere better. There is a booker on the contact page if you would rather skip the writing.

Read access to the existing system if there is one, thirty minutes each with two or three people who use it, and a named person who can make decisions. That is enough to write a scope. We do not need a specification — writing one is part of the work.

Yes, at a fixed price. It runs two weeks and ends with a written scope, a risk list, and a milestone price. It is deliberately standalone: you can take the document and build with anyone, including your own team.

Discovery usually starts within two to three weeks. Build slots depend on what is already in flight — we run at most three active projects, so if the next slot is two months out we will tell you that instead of starting badly.

Yes, before any detail is shared. We will read it rather than rubber-stamp it, and we will flag anything unusual, such as clauses claiming ownership of general engineering knowledge.

Discovery is a fixed price. Build work is priced per milestone, invoiced on delivery of each. Embedded and retainer engagements are monthly. We do not bill hourly for delivery work — it rewards the wrong things.

Most build engagements land between the price of two and six months of one senior engineer, depending on scope. That range is wide because the honest number depends on what discovery finds. You get a real figure at the end of discovery, before you commit to building.

A fixed-price milestone is fixed. If we misjudged the work, we absorb it. If you change what the milestone means, we re-scope it together and you see the new number before anything continues.

You do, from the first commit. Repositories, cloud accounts, domains, and pipelines are in your name. We keep no licence over the work and no leverage over your ability to leave.

Our engineers are not available as permanent hires through us. There is no no-poach clause on anyone we bring in as a specialist, and we will introduce you to good people if what you actually need is a permanent hire.

No. It misaligns the incentive to finish and hand over, which is the part of our model that works.

One hour a week for the demo, plus answers to questions inside two working days. Projects go wrong when a client cannot answer questions, so we would rather agree that up front than discover it in week six.

We are in Minato, Tokyo, and hold overlapping hours with Asia-Pacific, Europe, and North American mornings. Async by default — decisions get written down, so nobody has to be awake to unblock anyone.

English, Japanese, and Chinese, across scoping documents, code comments, and handover material — not only in meetings.

Yes, and it is usually the better outcome. Your engineers review our pull requests, we review theirs, and the handover is already half done by the time the project ends.

Milestone work stops at the end of the current milestone; you pay for what was delivered and keep everything. Monthly engagements need 30 days' notice. There is no termination fee.

A review period where your engineers drive and we comment. After that, a retainer if you want one, or nothing at all. Plenty of past clients are in the second category and we consider that a success.

TypeScript, Next.js, Node, and Postgres, because we are fast in it and it is easy to hire for. We will use something else when there is a reason, and we will tell you what that choice costs you in maintenance.

Usually. Rails, Django, Laravel, .NET, and Go are all fine. We will decline if the stack is genuinely unmaintainable and rewriting is not on the table, because we would be taking your money to make it worse.

Evaluation before agents. We pick a task somebody does by hand, measure the baseline, build the smallest thing that beats it, and put an evaluation suite in CI so a prompt change cannot silently regress. If a model is not the right tool, we build the boring version.

Into your infrastructure. For AI work we default to providers with no-training guarantees on business tiers, and we can run open models in your own cloud where the data cannot leave. Whatever we choose, it is written into the scope, not assumed.

Often, yes. It starts with a paid audit: what works, what is load-bearing, what has to be replaced, and what it costs. Sometimes the answer is that a rewrite is cheaper than the rescue, and we would rather find that out in week one.

We run the operating layer around expert contributors: sourcing and screening them against a written bar, calibrating them so two reviewers grade the same item the same way, writing rubrics against each platform's guidelines, and monitoring quality with gold sets and blind double-marking. The domain expertise usually comes from the client side; the process, measurement and reconciliation come from us.

Outlier, Snorkel AI, DataAnnotation.tech, Alignerr, Handshake AI, Pareto AI, Remotasks and Toloka, with engagements starting at different times rather than all at once. The services sheet has a chart of how that revenue built up and a short description of what each platform is.

We do interface design and design systems to a standard we are comfortable shipping. For brand identity and illustration we bring in people who are better at it than us, and you contract them directly.