Services
AI that runs on your data, on your infrastructure
Not a chatbot bolted onto your website. A system built on AWS, inside your own environment, against your own data, by somebody who understands the database underneath it.
The position, stated plainly
Most of what is sold as AI consultancy is a wrapper around somebody else's API with your data going through it.
I've built a production AI system on AWS services, running inside the client's own environment, on their own data, designed by somebody who has been working with databases since 1988.
If what you need is a chatbot on your website, buy one.
It'll cost you £30 a month and I'll tell you which one. Don't pay me consultancy rates for that.
If you need AI working against your own systems and data without shipping either to a third party, that's a different piece of engineering, and it's the one I do.
And I should be specific about the boundary: this is AWS.
Bedrock and the AWS data services around it, on your own account. That's where I've built it and that's where I know what the running costs, the access controls and the failure modes actually look like.
If your business runs on Azure or Google Cloud, somebody who works there every day will serve you better than I will, and I'd rather tell you now than learn it at your expense.
Who this is for
- You have data you can't send to a third-party API, because of contracts, regulation, or because it's the business
- You've been shown a demo that impressed you and you can't tell whether it would survive contact with your actual systems
- You want to know what this is genuinely useful for in your business, and you'd like the answer from somebody with no licence to sell you
The problem
Every company has now been shown a demo that looked extraordinary, and most have no way of telling whether it would survive contact with their own systems.
The demo works because somebody curated the data. Yours isn't curated. It's in four systems with inconsistent keys, three years of records where a field changed meaning halfway through, and a spreadsheet that turns out to be authoritative. Point a model at that and you get answers that are confidently wrong, which is worse than no answers at all.
Then there's the question nobody asks until the contract stage: where does the data actually go, who processes it, and what have you agreed to on your customers' behalf.
What matters, and what people skip
The interesting part of applied AI is almost never the model. It's the data layer: what you actually have, what state it's in, how it reaches the thing doing the work, what happens when it's wrong, and who can see it.
It's a database and infrastructure problem, and it's the part most AI consultancies are least equipped for.
What I do
- Production AI on AWS, in your own account, using Bedrock and the data services around it
- The data engineering underneath it, which is most of the work
- Retrieval against your own systems, with the access controls your business already has
- An honest assessment of whether a given idea is worth doing at all, including when it isn't
How it works
This almost always starts with an advisory session at £175 or a technical review, because the useful first question is whether the idea is viable against the data you've actually got. That answer is worth having before anybody quotes you for a build.
Build work is scoped first and then quoted as a piece of work rather than as time. Consultancy prices on this page exclude VAT.
What it runs on
- AWS
- Bedrock
- Linux
- MySQL
- MariaDB
- Aurora
- PHP
And your own infrastructure rather than mine.
Questions
The things people ask
Where does my data actually go?
To AWS, in your own account, in a region you choose. Nothing is sent to a third-party model provider unless you decide it should be, and that's a decision to take deliberately rather than by accident.
Is this worth doing for a company our size?
Sometimes. Frequently not yet. Book an advisory session and I'll tell you which, including when the answer is wait.
We already have a chatbot and it isn't very good.
That's usually a data problem rather than a model problem, and swapping models won't fix it.
How much does this cost to run?
Less than most people expect to build and more than most expect to operate. The running cost is the number to model properly, and it's the one vendors are vaguest about.
Can you work with our existing developers?
Yes. Often the right shape is that I do the data and infrastructure layer and they build the application on top.
What if we aren't on AWS?
Then I'm probably not your best option, and I'll say so rather than learn a second cloud on your budget.