Systems architecture & AI industrialisation
From POC to production.
Your AI prototypes work in the demo. Running them inside your information system — safely, under control and in compliance — is a different job. Thirty years of information systems architecture have taught me that.
- +30 yearsof information systems architecture
- Former CTOof iObeya, a SaaS software vendor
- TOGAF 9.1Certified enterprise architect
- Google CloudProfessional Cloud Architect
What I solve
Three situations. Three answers.
Companies are not short of AI ideas. They get stuck when it is time to scale — for lack of architecture, leadership or a compliance framework.
“Your POCs impress in demos, but none of them is in production.”
POC to Production
We start with a maturity assessment, pick the use case worth industrialising, then build an architecture that will hold: data, security, costs, operations.
“You need technical and AI leadership, not a full-time hire.”
Fractional CTO / CAIO
I take ownership of your technology and AI roadmap a few days a month: trade-offs, architecture decisions, steering teams and suppliers.
“The AI Act applies to your systems, and nobody really knows where you stand.”
AI Act Ready
Inventory of uses, risk classification, gaps to close, action plan: compliance becomes an engineering topic rather than a source of worry.
The method
What has been proven must hold in production.
Four steps, always in the same order. Each one produces a concrete deliverable, and each one can be the last if the value is not there.
Scope
Understand the business need, the existing landscape and the constraints. Pick the use case with measurable value.
Prove
A fast demonstrator, on your real data, with success criteria set before we start.
Architect
Design the target: integration with your systems, security, data sovereignty, control of inference costs.
Industrialise
Go to production, equip operations — observability, continuous evaluation — and transfer skills to your teams.
References
Organisations I have worked with
Organisations supported throughout my career, directly or through consulting firms.
Latest articles
All resources →The verdict: which architecture, for which haystack
Six ways to find one sentence in 33 novels, on one machine, compared.
RLM: the model that writes its own search
The library is a variable, the model writes Python, and it never calls itself.
Agentic search: let the model look
No index, four tools, and a model that decides where to look.
Let's talk about your next move to production.
45 minutes to understand your context and see whether — and how — I can help. No commitment, no sales pitch.
Book a scoping callPrefer to write? Get in touch