Insurance · SAMA · KSA
Najm Insurance
6,000+
Daily cases · 40+ cities
About
Koen Van Lysebetten
AI Architect & Governance Advisor · DevGap
I build the layer that ships regulated AI into production — constrained identities, attributable actions, human gates and kill switches for agents operating under real regulators. The model is the easy part. The plumbing is what kills you.
One conversation, no deck theatre · CV (PDF) for consulting & advisory engagements.
Twenty years of enterprise systems
I've watched the models change. The plumbing never did.
For two decades I've built and shipped enterprise systems — the unglamorous kind that have to stay up, stay auditable, and survive scrutiny. It started with Kapaza, the Belgian classifieds platform later acquired by Schibsted, and it led to leading regulated-AI delivery at DevGap across Europe, the Middle East and India.
The pattern repeats in every sector. Teams invest millions in models, data pipelines and proof-of-concepts — and almost nothing in governance, audit trails and kill switches. The pilots work in a controlled environment and become liabilities in production. The failure is almost never the AI. Nobody built the system around the model.
So that's what I do. I'm not here for the demo. I'm here for the part where an autonomous system makes a decision a regulator can question — and you have a traceable answer, a human in the loop, and an off switch that actually works.
The organizations that succeed with AI aren't the ones with the best models. They're the ones with the best plumbing.
The path here
From classifieds to regulated AI.
Early career
Kapaza → Schibsted
Built the Belgian classifieds platform that scaled to a national audience and was acquired by Schibsted — high-traffic systems where reliability was the product.
Then
Enterprise delivery · DevGap
Twenty years of building systems that have to stay up and stay auditable — across banking, insurance, government, transit and hospitality intelligence.
Now
Regulated-AI delivery · EU · ME · India
Leading governance-first AI engineering: the Trust Engine — four non-negotiables that decide whether an agentic system ships or dies. Four are in production.
Government · India
Govt of India / NMML
€10M+
180+ endpoints · 99.9% SLA
Public transit · Belgium
De Lijn
5,000+
FTE in scope · board-approved
Hospitality intel · USA
US Restaurant Intelligence
200 → 3
FTE to agents · ~90% cost cut
Consulting & advisory
Who I'm engaged for.
Three rooms, one job: make the autonomous system survive the regulator and actually ship. The CV (PDF) above has the full engagement history.
The CEO
You approved four AI pilots. None are in production. The board wants ROI, and the EU AI Act deadline is real.
I build the system that survives the regulator and delivers the projected return.
The CTO
You inherited AI point solutions with zero governance. You need a reference architecture to get CISO and Legal sign-off — now.
I provide the ADR patterns, constrained-identity models and observability stack.
The Investor
You're evaluating AI scale-ups. You need to know whether the "moat" is true infrastructure or a thin model wrapper.
I run technical due diligence on the integration, governance and defensibility layers.
The work, concretely
The Stack Bridge.
What changes, layer by layer, when a system is built governance-first instead of demo-first.
Layer
Traditional AI
The AI Plumber stack
Identity
Shared API keys, "god-mode" access
Constrained service accounts, cryptographic verification
Execution
Synchronous, unbridled loops
Asynchronous, human-in-the-loop authorization gates
Observability
Black-box models, console prints
Fully reversible reasoning trace on a read-only ledger
Safety
Vague prompt guidelines & hope
Hard kill-switch thresholds, automated suspension
How I work
Evidence over adjectives.
Instructions are a wish. The keyring is a guardrail.
Safety doesn't come from telling an agent to behave. It comes from least-privilege identities that make the dangerous action impossible in the first place.
Intelligence is rented. Piping is owned.
Models are commodities you swap out. The governance layer around them is the durable asset — and the thing a regulator actually examines.
No deck theatre.
The cases do the persuading. One honest conversation about your system beats a slide on the future of AI every time.
Useful before it's finished.
A 4-page briefing a board can act on is worth more than a perfect document nobody reads. Ship the artifact that changes the decision.
Koen Van Lysebetten
AI Architect & Governance Advisor · DevGap
Read the book →