AI services
A model that knows your business, not the internet’s.
A general model is fluent about everything and expert in nothing you sell. We take what your business already knows — its documents, its tickets, its history — and turn it into a model that answers like someone who works there.
What we actually build
We do not train a model from nothing. Nobody sensible does. We start from the best model that already exists for the job and make it yours.
Adapt an open model
Fine-tuned on your data and run on hardware you or we control. The weights are yours, the model runs where you say, and nothing about your business leaves for a third party’s API.
Optimise a frontier model
When the best answer needs the best model, we put your knowledge around it instead of inside it — retrieval, tools and grounding tuned until it stops guessing and starts citing.
Make the data usable
Most of this work is not modelling. It is finding what a business knows, getting it out of PDFs and inboxes, and deciding what is true when two sources disagree.
How an engagement runs
Three steps, and you can stop after the first one with something worth having.
Step one
Diagnostic
Two weeks, fixed price. We look at what data you hold and tell you which of your problems a model can actually solve — and which ones are a spreadsheet, a rule, or a process change. You keep the report either way.
Step two
Build and evaluate
We build against a test set drawn from your real questions, agreed before we start. You see the score, not a demo. A model that is right 70% of the time is a different product from one that is right 95% of the time, and you should know which you have.
Step three
Run it, or take it
We host it and run it for you, or you take the weights and the pipeline in-house. Both are real options, and we will not hold your model hostage to a subscription.
Why us
We run this on ourselves first
Everything on this page is how we built our own platform. The residency controls, the grounding that makes an agent say “we do not sell that”, the record of what leaves for inference — none of it is theory we would try out on you.
- Your data trains your model. It never trains anyone else’s.
- Training and inference in the region you choose, with every call recorded against it
- A record of what categories of personal data reach a model, and from where
- Evaluated against your questions, not a public benchmark
Where this fits
AI products is the ready-made thing: agents that answer email, WhatsApp and web chat from a catalogue you type in. It works on day one and costs a subscription.
AI services is the bespoke thing: your own data, your own model, your own deployment. It takes weeks and costs a project fee.
Most businesses should start with the product and come here when they hit its edges — when the answer needs knowledge no catalogue holds. We will tell you which one you need, including when it is the cheaper one.
Bring us your hardest question
The one your team answers by asking the person who has been there longest. If a model can learn that, it is worth building. If it cannot, we will say so.