What we mean by AI.
Products that use language models to read documents and answer questions about their contents, hold a conversation with a user, extract structured data from sources that are not structured, generate text and images, and automate manual processes with agentic workflows.
The hard part is not the model, which belongs to someone else and can change. It is what surrounds it: how documents arrive, what happens when an answer is wrong, and how to trace what produced it.
A prototype that answers the demonstrator's questions well is within anyone's reach. The rest is not.
The technologies we work with
The technologies we work with
When we do not use it
When we do not use it
When the rules are known and can be written down. If a decision depends on criteria someone can articulate, a rule costs less, is faster, makes fewer mistakes — and can be corrected knowing exactly what is being corrected.
And when there is no way to verify the answer. A system that produces plausible outputs in a context where no one can check them shifts the risk onto the client.
What we have used it for
What we have used it for
BuildMarketplace
Sailsquare — 5 years on the platform, alongside their team.
5 yearsActive
[REDACTED]
ScaleApplied AI company
From prototype to production product.
2 yearsActive
The way we work is the same across technologies: peer review before release, separate environments, and automated pipeline checks. How we work