Claims come with methodology
Any performance or cost figure we give you arrives with the method used to produce it, so you can check it or reproduce it. Estimates are labelled as estimates.
Approach
Four stages.
You can stop after any one.
Most AI projects fail late and expensively. Ours are structured so the expensive parts come after the evidence, not before it — and so you keep everything produced at each stage regardless of what you decide next.
Engagement model
How the work is sequenced
STAGE 01
Scope
We map the problem, the data available, and the constraints you are working within — technical, regulatory, and budgetary. The deliverable is a written assessment that says what is worth doing, what it will roughly cost, and what the risks are. If the honest answer is that AI is the wrong tool for this problem, that is what the assessment says.
STAGE 02
Prove
A working prototype measured against your real data, with a benchmark you can run yourself. This stage exists because impressive demos and production-grade results are different things, and the gap between them is where budgets disappear. You get the prototype, the evaluation harness, and the numbers — whether or not the result is favourable.
STAGE 03
Build
Production implementation: training or fine-tuning as required, deployment, integration with your existing systems, and the operational tooling around it — monitoring, logging, and guardrails. Documentation is written as the work happens rather than reconstructed at the end.
STAGE 04
Operate or hand over
We run the system for you on dedicated capacity, or we train your team and hand over everything — code, configuration, runbooks, and the reasoning behind each decision. This choice is made at the end, when you know what operating it actually involves, rather than committed to at the start.
Working principles
What you can expect
You keep what we build
Code, model weights, evaluation harnesses, and documentation belong to you. Nothing is locked behind a platform you would have to keep paying for to retain access.
We will tell you not to build it
A recommendation against the project is a legitimate outcome of the scoping stage, and one we have no financial reason to avoid — we do not resell anyone's platform.
Your data stays yours
We work with open-weight models so your data never has to cross into a third-party API. It is not used for training, not retained, and not shared.
Fixed scope before fixed price
Each stage is defined and quoted before it begins. Scope changes are discussed and agreed rather than absorbed silently into a timeline.
Built to be maintained without us
Systems are documented and structured so your team can operate and extend them. Ongoing involvement should be a choice you make, not a dependency you inherit.
Start with a scoping conversation.
Short, direct, and enough to tell whether there is a project here worth running.