Journal

Notes from the bench.

On building and running AI systems for businesses. Published when there is something worth reading, not on a schedule.

Questions we get asked

Short answers to the questions that come up first.

Do we need our data ready before we start?

No. The assessment exists to find out what you have and what is missing. A first project usually starts with data scattered across tools, so the plan is built around that rather than waiting for it.

Which model will you use?

The smallest one that passes your evaluation set. Hosted and open models are tested against the same cases and the choice is made on evidence, not on brand. It is written down so you can revisit it.

Can it run inside our own cloud?

Yes, and usually it should. Deployment goes into your account, under your access controls, so the system and its data stay yours. Studio access can be removed the day you decide.

How will we know it works after launch?

Every answer passes rules before anyone sees it, and a weekly sample is scored against your evaluation set. A monthly report shows the trend in plain language, incidents included.

What happens if we stop working with you?

You keep everything: the code, the evaluation set, the runbook and the infrastructure. The runbook is written so an engineer who has never met us can operate the system.

How much does it cost?

Each phase is priced on its own before it starts, so you can stop at any door. Use the scope tool for the shape and the weeks, and write to us for a number.