Questions we get asked first

Frequently asked questions

Can you keep running it after handover, or is it ours to operate?

Either. The default is that everything transfers on final payment, with documentation and a training session, and your team runs it. If you would rather not carry it, we keep it on a monthly arrangement: scheduled evaluation runs, prompt and model upkeep, a named response time during agreed hours, hosting in our accounts or in yours with us on call, and feature work against an allocation you approve. Ownership does not change in either case, and you can switch from one to the other.

We do not have clean data. Can you still do anything useful?

Usually, though the first honest finding is often that the documents need work before the model does. We start with a source audit: what exists, how current it is, and how much of it is contradictory. Sometimes the answer is a smaller system over the material that is actually reliable, which is cheaper and works better than a large one over everything.

How do we know the AI is not making things up?

Two things. Retrieval returns a citation, so any answer traces back to the document it came from and your staff can check it. And we build an evaluation set from your own queries and run it on every change, so accuracy is a number on a chart. Where something genuinely cannot be measured, we will tell you that instead of reassuring you.

Which model do you use?

Whichever is currently best for the task and the budget. We build so that changing it later is a config change, because in a market moving this fast, tying you to one provider is not our decision to make.

Does our data get sent to train someone's model?

Not without you deciding it should. Enterprise API tiers from the major providers exclude your data from training by default, and we set that up as part of the build. Where the material is sensitive enough that leaving your infrastructure is the problem, we will scope a self-hosted option and be straight about what it costs.

What does a project cost?

The first meeting is free and is for getting to know you, the business context and the high-level requirements. If we both agree to continue, discovery begins at USD 150 and includes a proposed solution with up to two feedback iterations. Further iterations are billed at the agreed hourly rate. After that, most fixed-scope builds run six to eight weeks and are quoted against a written scope. Payment is 40% to commence, 30% on design approval, and 30% on completion. Anything larger we would rather break into phases you can stop after.

Can our own team maintain it afterwards?

That is the intended outcome. Code, prompts, evaluation sets and infrastructure transfer to you on final payment, with documentation and a training session. If you want to keep going independently, setting our working method up inside your team is the version of this work we most like doing.

Do you still build ordinary software?

Yes. Web platforms, integrations and internal tools are a good part of what we do, and an AI feature bolted onto a business that cannot deploy reliably is not an AI problem. Quite often the useful first engagement has no model in it at all.