Approach
The last furlong is where it's decided.
What we think the work actually is, and — more usefully — what we won't pretend to be.
Why the name
A furlong is an eighth of a mile. On a racecourse, posts count down the distance remaining, and the last furlong is where the result is settled — not because the earlier ground didn't matter, but because that's where it's decided.
The parallel is close enough to be useful. Choosing a model, writing prompts and reading benchmarks is the first seven furlongs: necessary, visible, and where most of the industry's attention goes. Deployment, integration and the security review are the last one. That's the ground we work on.
What we believe
The boundary is the design
Where your data is permitted to go determines the architecture. Start there and the rest follows. Start with the model and you will redesign it later, after someone in security has read the proposal.
Owning the stack beats reselling one
Because we write the software, we can change how it deploys to fit your constraints. A reseller facing the same constraint can only change the invoice.
Specifics, or we don't understand it yet
If we can't tell you which host, which protocol and which port, we haven't understood your problem well enough to be paid for it. Adjectives are not a design.
Small is a trade, not a virtue
Being a boutique means you get people who know your deployment and answer directly. It also has costs. We'd rather name them than let you discover them.
What we don't do
Shorter engagements and better outcomes come from ruling things out early. This list is here so you can disqualify us in two minutes rather than three meetings.
We don't train or fine-tune foundation models.
We integrate them. If your problem genuinely needs a custom-trained model, you want a research group, not us.
We don't resell a hosted wrapper and call it a platform.
If a hosted product would genuinely serve you better, we'll name it and you can go and buy it.
We don't claim certifications we haven't earned.
You won't find compliance badges on this site. If your procurement requires an audit report we don't hold, we'll tell you on the first call rather than consume a review cycle finding out.
We won't recommend AI for a problem that doesn't need it.
A fair amount of what arrives described as an AI project is a reporting problem or a data-quality problem in a costume. Saying so is cheaper for both of us.
We're small.
If you need a large delivery team on site, we are the wrong firm. Where you need continuity guarantees, documented handover comes as standard, and escrow can be arranged on request.
Finish
Still the right fit?
Then the useful next step is a conversation about where your data is allowed to go.