Four regulated systems, shipped.
Each one anchors a pipe in the method. These are delivered systems under real regulators — not pilots, not slideware. Figures are as delivered; attribution never moves between cases.
A vision-AI claims system operating at national scale, where one over-privileged agent could breach the strictest financial regulator in the Gulf.
Najm processes thousands of motor-accident claims a day across Saudi Arabia, and wanted vision AI in that pipeline — automated damage assessment at national scale. The governance gap: the platform was a monolith with broad database access, so dropping an autonomous agent into a SAMA-regulated environment meant any over-privileged process could reach the wrong record. At that scale, an agent acting without a trace isn’t a bug — it’s a reportable breach.
6,000 claims a day, under the Gulf’s strictest financial regulator — and not one agent with admin rights.
Your exposure was never the model. It’s the service account behind it — and one agent with broad write access is one audit finding from a breach.
Claims, underwriting, KYC, fraud — the moment an agent can reach a system of record, the regulator’s question is no longer “is it accurate?” but “who authorised that, and can you prove it?” We scoped every identity to its blast radius before a single model went live. The AI Plumber Day maps your agent identities to your regulator’s requirements in a single day.
A sovereign cultural estate where every automated action must be attributable — and survive the scrutiny of a national government.
The Government of India commissioned an end-to-end immersive AV and AR/VR estate for the Prime Ministers’ Museum — 180+ endpoints synchronised from a central control engine. The governance gap: the infrastructure cannot leave the country, and at ministry scale “we don’t know why the system did that” is not an answer you give a national government. Attribution and sovereignty weren’t features — they were the terms of the contract.
180+ endpoints under a national government — every automated action traceable to a single ledger entry in under 30 seconds.
The question isn’t whether your AI works. It’s whether you can explain every action to a minister, a court, or an FOI request — and whether the data ever left the country.
Most public-sector AI can answer neither. We built the audit trail to survive a regulator and ran the whole estate on sovereign infrastructure, so attribution and residency were architectural, not promised. A Board Briefing gives your leadership a decision-ready governance posture for public-sector AI — strategy, risk and a proposal they can sign.
A 5,000-person public body adopting AI without handing irreversible decisions to a system no human signed off on.
De Lijn — Flanders’ public transport operator, 5,000+ staff — wanted to move on AI across the organisation. The governance gap: under the EU AI Act and GDPR, a public body cannot hand irreversible decisions to a system no human signed off on. But blanket caution kills the velocity that makes AI worth adopting. They needed both — defensibility and speed — and a posture leadership could put their name to.
A 5,000-person transit authority got a board-approved AI roadmap — EU AI Act-classified, 129% projected ROI.
The EU AI Act isn’t a checkbox you bolt on later. It decides what you’re allowed to build at all — and classifying after you’ve built means you rebuild.
Transit, utilities, public services — get the classification right first and you ship with the board’s name on it; get it wrong and you’re unwinding a system the regulator won’t accept. We classified every use case up front, then placed human gates only where the Act requires them. A Board Briefing turns your EU AI Act exposure into a decision your board can actually sign.
A multi-agent pipeline with real autonomy — and a hard backstop that fires before a runaway loop becomes a runaway bill.
A US hospitality-intelligence operator ran a 200-person manual operation gathering and structuring restaurant data, and wanted to replace it with a LangGraph multi-agent system. The governance gap: unattended agents loop, spike cost and drift at 3 AM with nobody watching. At multi-agent scale, a single runaway loop can burn a month’s budget before anyone logs in — the economics only work if failure is contained automatically.
200 people replaced by 3 agents at ~90% lower cost — with a kill threshold that fires before a runaway loop becomes a runaway bill.
The upside of agents is obvious. The downside is invisible until the invoice arrives — and autonomy without a backstop isn’t efficiency, it’s an unmonitored liability that happens to be cheap until the night it isn’t.
Back-office automation, data ops, customer workflows — the agents that save you 90% are the same ones that can spike spend or drift at 3 AM. We gave them real autonomy inside a boundary they can’t cross, with telemetry that suspends them on breach. The Build Sprint takes one of your processes from manual to a governed agentic prototype — on your data — in two days.