AI development
AI development
AI features built into a product people already use — retrieval over your own documents, agents that take actions, and the evaluation work that tells you whether any of it is right.
From £75/hrAn estimate before any work startsOne developer, start to finish
01
Most AI projects do not fail at the model. They fail because nobody defined what a correct answer looks like.
Without that definition there is no way to tell whether the thing works, and no way to notice when it stops. Every build here ships with a test set and a score you can run again after any change — which is also what makes it maintainable by someone who is not me.
02
What actually gets built
Retrieval over your own content
Your documents, tickets, contracts or product data, indexed so a model answers from them rather than from its training — and refuses when it has not found them.
Agents that take actions
A model that can call your own APIs: raise the ticket, draft the reply, move the record. With the guardrails that decide what runs unsupervised and what waits for a human.
Classification and extraction
Reading invoices, routing enquiries, pulling structured fields out of free text. The unglamorous one, and the one that pays for itself fastest.
The evaluation set
Real inputs, the answers they should produce, and a score. It outlives the model, the vendor and the developer, and it is what turns prompt changes into ordinary engineering.
03
How the work runs
You describe it
Two sentences is enough to start. What you are building, who it is for, and where it is stuck if it already exists.
A scope and a price, inside 24 hours
Not a range with a shrug attached: what is in, what is deliberately out, what each stage costs and what it depends on. If something cannot be priced honestly yet, that is said rather than padded.
Built in the open
Your repository from the first commit, deployed somewhere you can see it from the first week. Progress is a working thing you can open, not a status call.
Handed over properly
Your accounts, your keys, your infrastructure, and a README that assumes the reader is not me. Ongoing work can run on a retainer; nothing breaks if it does not.
- Rate
- From £75/hr
- Scope and price
- Inside 24 hours
- Written by
- The person who scoped it
Not sure which of those you need? Describe the problem in two sentences and you get back a scope, a price and an honest view of whether AI is even the right tool for it.
Start a project →Common questions
How long does an AI feature take to build?
A narrow, well-defined one — retrieval over a document set, or an extraction pipeline — is usually a matter of weeks rather than months. The variable is almost never the model work; it is how clean and reachable your data is, and how long it takes to agree what a correct answer looks like.
Do we need our own model, or fine-tuning?
Almost never to begin with. Retrieval plus a good prompt and a proper evaluation set beats a fine-tuned model in most business cases, and it is far cheaper to change. Fine-tuning earns its place when you have a large set of examples of the exact behaviour you want and prompting has genuinely run out of room.
Where does our data go?
Wherever you decide, and it is a decision made before any code is written: which provider, which region, whether anything is retained, and what the data processing terms actually say. If you have obligations that constrain this, they get designed in at the start rather than discovered at the security review.
04
Tell me what you want it to do
Two sentences is plenty. You get back a scope, a price and the shortest honest route to it — usually the same day, from the person who would do the work.