What we’ve built
We build the AI systems most people only advise on.
Most AI advice comes from people who’ve never shipped one. Everything on this page we designed, built, and deployed ourselves — including a product that’s live to the public today, with real payment processing built in. Proof, not prototypes.
The Work Bank is public work we analyse. This is what we’ve built and run ourselves.
We don’t just advise on AI. We build production AI systems — and we use them to solve the problems keeping operators up at night.
Could your AI pass an independent governance audit?
We’ve built and deployed the governance layer most AI projects bolt on too late — enforced guardrails on what the model can and can’t do, consent and erasure surfaces that honour GDPR’s right to be forgotten, transparency disclosures aligned to the EU AI Act, per-user data isolation, and an audit trail behind every action. In a separate client build, we demonstrated confidential-data egress control: market-sensitive material that can never surface in external-facing output.
We make your AI audit-ready. We assess what you’re running against these same controls, close the gaps, and give you the evidence trail that turns “we think it’s fine” into “here’s how it decides, and who’s accountable.” The readiness is the offer — backed by controls we’ve already built, not a framework drawn up for the pitch.
Is your business agent-ready?
Systems built to work with other systems. Multi-provider model orchestration that routes each task to the right model, fails over automatically when one degrades, and keeps cost under control — plus live integrations into external services and a live payment rail, running now in a product that’s open to the public. The unglamorous engineering that lets software join automated workflows instead of sitting outside them.
We get you agent-ready. AI agents are starting to discover, compare, and transact on behalf of the people you sell to — and they route around anyone whose data they can’t read. We make your product and operational data machine-readable and structured, connect your systems to agent workflows and the emerging commerce and payment protocols, and position you to be found and used by agents rather than bypassed. In e-commerce that’s the difference between being recommended and being invisible; in operations, it’s whether your systems can join an automated workflow at all.
Does your AI actually remember your business?
Most deployed AI is stateless — it forgets between sessions and never builds real memory of your operation. We’ve built and deployed a semantic-memory system that does: vector-based retrieval (pgvector), memory held in tiers — canonical, stable, episodic — and a fact lifecycle that lets detail decay when it should, so the right context surfaces at the moment of generation and stale context doesn’t.
We give your business durable AI memory — a system that genuinely remembers your people, your decisions, and your operational context, instead of restarting from zero every conversation. Governed by the same controls above: durable where it helps, private by design.
This is the engineering behind the advice. When we scope your bottleneck, this is the depth we bring — and the proof that we’ve built the thing, not just talked about it.