Questions, answered plainly.

We're an AI development services firm. We design, build and run AI systems for businesses: agents that take action inside your tools, assistants that answer from your own data, and models tuned to your problem, from first prototype to production.

With a discovery call. We map the workflow, the data involved and what “good” looks like for you. If AI is a fit, we propose a first step, often a short strategy sprint or a prototype on your real data. If it isn't a fit, we'll tell you.

Through a shared channel, regular check-ins and written updates you can read on your own time. You see working software early, and changes to scope are agreed before we act on them.

Yes. We add AI features to the web and mobile apps you already run, or build new AI-first products end to end. We start from your stack and your users, not from a rewrite.

We scope before we price. After the discovery call we propose one of three shapes: a strategy sprint with a fixed length, a scoped build with milestones, or an ongoing run-and-improve arrangement. Each proposal states what's included, how we'll measure it and what it costs.

It depends on the workflow and the state of your data, so we don't guess at a timeline before discovery. The strategy sprint runs two weeks and ends with a working prototype; for a build, the proposal sets out the milestones and dates we commit to.

The payment schedule is set in the statement of work for each project and is usually tied to milestones. Our standard terms are in the terms of service.

An AI system has running costs, mostly model usage and hosting. We estimate them during the prototype, measure them in evaluation alongside accuracy and speed, and design to the budget you set.

Your data stays yours. We prefer private deployments in your own environment or cloud account, give access only to the people working on the project, and never train models on your data without your consent.

Often, with the right setup. We ask where the data must live, who may see it and which rules apply, then design around those limits: private deployment, redaction, access controls and logging. If a requirement rules out a design, we say so up front.

Yes. Without an NDA, our terms let us show completed work in our portfolio; with one in place, we don't.

Whichever fits the job. We're model-agnostic: we compare hosted and open models on your task, weighing accuracy, cost, speed and where the data may go, and we build so the model can be switched when a better one arrives.

Usually not at first. Most work starts with a strong existing model, your data and good retrieval. When that isn't enough, we fine-tune or distil a model for your task and budget, and the evaluation shows whether it was worth it.

We agree the bar with you before we build, then turn it into a test suite of real examples with expected results. Every version is measured against it for accuracy, cost and speed, and you can read the results yourself.

It will, sometimes, so we design for it. Systems cite their sources where they can and hand off to a person when it matters. Mistakes found in production become new test cases, so they stay fixed.

You do. Once the project is paid in full, the custom code and deliverables built for it, such as prompts and evaluation suites, are transferred to you, and your data was never ours. We keep the general-purpose tools and reusable components we develop independently.

If you want us to, we keep running it: we watch the system in production, review its evaluations and failures, and keep improving it against the same bar. If you'd rather run it yourselves, we hand over what your team needs: code, documentation, evaluations and a walkthrough.

Yes, as part of a run-and-improve arrangement. What it covers, from monitoring and fixes to new evaluations and model updates, is agreed in writing for each system.