Software · Applied AI · Industrial systemsApplied AI · International
An engineering studio for applied artificial intelligence — autonomous agent architectures, full-stack platforms and the infrastructure underneath them. We run models on your own hardware, wire them into real processes, and make every output traceable to its source.
What we do
Most of the market sells pilots. We build the layer underneath — infrastructure, controls and evidence — so the thing survives contact with a real business.
Autonomous agents that do process work end to end — read, extract, compare, draft — with permission gating, independent verification and a hash-chained record of every decision. Draft-only by design: nothing sends, signs or pays without a named human.
Models running on your own hardware or your own cloud tenancy, so confidential data never leaves the perimeter. Residency by architecture rather than by contract — enforced in the runtime, not promised in a policy.
Production platforms built and operated end to end — application, data layer, deployment and the monitoring that tells you it's still working at three in the morning.
Technology implemented inside physical production — instrumentation of additive manufacturing processes, machine data captured and made useful, and AI applied where the process actually runs rather than where it's easy to demo.
Complete custom systems from the board up: we design the electronics, then write everything that runs on them — firmware on the metal, embedded Linux underneath, applications above.
The part that decides whether any of it reaches production in a regulated company. Management systems, data-protection architecture and AI Act classification, designed in from the start rather than retrofitted.
Selected work
Systems in production, shown anonymized. Client names and live links are shared on request.
Applied AI · Enterprise
A multi-agent system running back-office process work for an organisation — document mapping, extraction and drafting, with every figure computed in code, every claim carrying a source, and a verification agent reviewing output before any human sees it.
Shown anonymized · details on request
Software · Full-stack
A complete trading platform built from the ground up — order matching, ledger, accounts and live market data, production-hardened on PostgreSQL. Proof the studio's software goes well beyond integrations.
Shown anonymized · details on request
Embedded · Firmware
A bespoke real-time control system for live performance — custom board, from-scratch firmware, sub-frame audio response and field-tunable configuration. Built for a private client; details kept confidential.
Shown anonymized · details on request
Applied AI · Education
A fully agentic authoring layer wired straight into Moodle — agents draft, structure and produce entire courses, then publish them into the client's LMS as native content. Nothing goes live without human sign-off. Shown anonymized at the client's request.
Shown anonymized · details on request
How a project works
Built for work that has to survive production. You talk to the people building it — not a sales desk.
We start with what exists — the process, the data, the systems and the constraints. Most of what matters is never written down, so we go and look.
A bounded slice, on your real data, with the success and stop criteria written down before we start. Measured, not asserted.
The full system, with the controls in the runtime rather than in a document — access gating, independent verification, and a record you can audit afterwards.
Infrastructure as code, open-weight models, configuration handed across. You can end the engagement and keep a running system — that's a property of the design, not a promise.
Why us
Software, infrastructure and governance under one roof — which is why the work that has to hold up under scrutiny ends up here.
We didn't make the AI trustworthy by trusting it more. Every number is computed in code, every claim carries its source, and anything the system didn't know shows up as a gap instead of an invention.
Confidential work runs on infrastructure you control, enforced at the network layer. Not a policy commitment — a rule that a compromised process still cannot get around.
The work nobody else will quote: legacy estates, merged systems, processes that only exist in someone's head. If it can be mapped, we find the way.
No lock-in by construction. Open weights, infrastructure as code, documentation that assumes we won't be here. The measure of the work is that it survives us.
Start a project
Tell us what you're trying to do — a process that eats people's weeks, a system nobody can explain, or an AI project that stalled at the pilot. We'll come back with what's actually possible, how we'd prove it, and what it costs.