Services · Build
Software that survives contact with production
Prototypes are easy and demos are easier. We build the version that handles the awkward data, the integration nobody documented and the Monday morning after launch — and then we keep running it.
Talk about a buildHow a build goes
Five stages, and you can stop after any of them
Each one produces something you own and could hand to somebody else. That is deliberate — a partner you cannot leave is not a partner.
We work out what the thing actually has to do, which of your existing systems it must live with, and what would make it a failure. You get a written problem statement, the constraints we found, and an honest view of whether this should be built at all.
A working slice through the riskiest part, running on real data where we can get it. Not a clickable mockup — code you can use, deliberately narrow, built to find out whether the idea holds before the budget is committed.
Fortnightly increments into an environment you can reach whenever you like. Every change goes through review and an automated gate: types, tests, security advisories, and a deploy that proves itself healthy before it takes traffic.
Load behaviour, failure modes, backup and restore actually restored, access control, audit trails, and the runbook for the night something breaks. This is the stage most projects discover they skipped.
Monitoring, patching, incident response and continued delivery. We keep what we build, which changes how we build it — nobody writes a clever shortcut into a system they will be paged about.
What we build
The work that comes to us
AI-powered applications
Most commonRetrieval-grounded assistants that cite their sources, document and data extraction, classification, and agent pipelines with a human in front of anything consequential.
Internal platforms
The systems that run an operation — scheduling, workflow, approvals, reporting. Usually replacing a spreadsheet that grew a following, sometimes replacing a product that stopped fitting.
Integration and modernisation
Legacy estates read, documented and translated. Business logic recovered from code nobody has owned for years, and a route to a platform you can keep building on.
Data and analytics services
Pipelines, warehouses and the reporting layer people actually open. Including the unglamorous part where the numbers reconcile with the system of record.
How we work
Commitments, not values
These are things you can hold us to during an engagement, which is a different category from words on a wall.
You own everything
Code, infrastructure definitions, documentation and credentials, in your accounts and your repositories from the first commit. No handover event, because there is nothing being held back.
Estimates come with their assumptions
A number without the assumptions under it is a guess dressed up. When an assumption breaks we tell you the same week, not at the end.
We will talk you out of it
If the problem is better solved by configuring something you already own, we will say so. That has lost us work and we would still rather say it.
AI where it earns its place
Used when it beats the deterministic alternative, and not otherwise. Where it is used, it shows its reasoning and a person approves anything that matters.
Tests and types are not optional extras
They are how a system stays changeable in year three. They are in the estimate rather than in a later phase that never arrives.
One team, one thread
The engineers doing the work are in the conversation with you. No account layer relaying questions to people you never meet.
Describe the problem, not the solution you had in mind
A short conversation usually establishes whether this is a build, a configuration job, or something you should not do at all.