Services

AI systems: build and run

I design, build and implement AI systems for businesses, SMEs, governments and institutions. The work connects AI agents, workflow automation and decision support to the systems your organization already uses, with evaluation, oversight and a clear operating responsibility.

Who it is for

Business owners improving a specific operation, enterprise teams integrating AI across functions, and government leaders developing institutional capabilities. Engagements can address one defined workflow or a broader implementation, with scope agreed around your priorities and resources.

What you get

• AI agents and multi-agent systems that act reliably within set boundaries
• Workflow automation across real operations
• Retrieval and knowledge systems grounded in your data
• Evaluation and quality-assessment harnesses so you can trust the output
• Production deployment, monitoring, and the governance around it

How it works

1. Fix the right problem before building
2. Build to production standard, not slideware
3. Adversarially verify before anything ships
4. Deploy, monitor, and run

Questions

Common questions

Can AI actually handle our WhatsApp messages, calls, and orders and put them into our system, or is it too early?

It works today when the scope is drawn properly. The reliable pattern is that the system handles repetitive traffic, prices, availability, hours, order status, and booking, and routes anything unusual to a person with the context attached. What breaks projects is asking one assistant to do everything on day one. Language matters here too, and I build for Arabic, French, and English. Integration into the system you already run is usually the real work, not the conversation itself.

Should we build the AI agent into our product ourselves with an API, or is that a mistake for a small team?

For a small team, build the thin part yourselves and buy the rest. Owning the prompt logic and the product surface keeps you fast. Rebuilding infrastructure that already exists spends the runway you need for customers. The decision I would test first is the failure path: what your agent does when it does not know, and how a person takes over cleanly. Teams that skip that end up rewriting everything after their first bad week in production.

Is it smarter to buy an off the shelf platform or have someone build a custom system for us?

Buy when your process looks like everyone else's and the vendor's roadmap points where you are going. Build when the workflow is your advantage, when the integration is the hard part, or when your data cannot leave your environment. Most institutions end up with both, and the value sits in the joins between them. I will tell you when a product already on the market does the job, because paying me to rebuild it would waste your money.

What is the cheapest way to run AI agents without burning our budget on usage costs?

Design for cost the way you design for speed. Route easy traffic to small models and reserve the expensive ones for hard cases, cache what repeats, keep prompts short, and cap what any single request can spend. Measure cost per resolved task rather than cost per call, because that is the number that decides whether the system pays for itself. Self hosting saves money only at real volume, and it moves the cost into engineering time you may not have.

Who runs the system after it is built?

Whoever you decide, but decide before launch. I build and run multi-agent systems in production, so I can operate it, or set it up so your team can, with the monitoring, escalation rules, and documentation to do that safely. An AI system is not a website. It drifts as your business changes and it needs someone watching what it gets wrong. Any proposal that ends at handover has delivered half the job and left you the harder half.

Related: Governing AI that acts · The Crucible verification method · AI governance & compliance · All services