Services · AI and agentic systems

AI features and agents that ship, with costs and failure modes under control.

I build AI into real products: provider-agnostic AI layers, assistants that act only on confirmation, multi-agent workflows with logging of every call, and AI features inside existing software. The aim is useful behaviour you can trust, not a demo.

Book this →All services

Who it is for

  • Products that want AI features without locking into one model provider
  • Teams that need an internal assistant reading their own business data safely
  • Companies exploring agents and wanting a design that fails safely
  • Regulated businesses that need AI switched off, local or provider-approved

How it is engaged

FormatScoped build or a short discovery sprint
Typical first stepA one-to-two week discovery that ends in a working prototype on your data
ProvidersAnthropic Claude, OpenAI and others behind one layer
OwnershipPrompts, code and evaluation data are yours

What you get

Use-case and risk review
Which tasks AI should and should not do in your business, with a cost estimate.
Provider-agnostic AI layer
One interface over several model providers, with health checks and fallbacks.
Assistants on your data
Permission-aware assistants that read only what the user may see and propose actions for confirmation.
Agent workflows
Role-specialised agents with their own prompts, token and latency logging, and human escalation as a designed state.
Evaluation and cost tracking
Measurable checks on output quality and a per-call cost record.
AI policy
Off, local model or approved provider, configurable per customer.

Proof: the work behind it

Real engagements and builds, each written up from the code and the running system.

AedahSuite: twenty-two business apps for UAE and GCC companies on one platform
AedahSuite: twenty-two business apps for UAE and GCC companies on one platform
aedahsuite.com
Small and mid-sized companies in the UAE usually run on a patchwork: one tool for the books, a payroll bureau for WPS, a CRM that does not know about…
D.U.N.E: eight specialised AI agents working one project
D.U.N.E: eight specialised AI agents working one project
dune-self.vercel.app
D.U.N.E stands for Digital Universe for Nomadic Excellence. It is an internal AI workspace for the Nomad Universe group, and it is not a chatbot. A…
Nokhba.ai: an AI career platform for the GCC, built and run end to end
Nokhba.ai: an AI career platform for the GCC, built and run end to end
nokhba.ai
Nokhba (نخبة, "the elite") is a career platform for the Gulf and wider MENA market: Arabic-first with a full English experience, and switchable…
AedahSocial: an AI-first social media suite, designed screen by screen
AedahSocial: an AI-first social media suite, designed screen by screen
Case study
Most social media management tools added AI as a button. AedahSocial was designed the other way round: every feature assumes AI is doing the first…

Questions

Will our data be used to train a model?

Not through anything I build. Providers are chosen and configured so business data is not used for training, and the AI can run on a local model where that matters.

Can the AI change our records by itself?

Only if you design it to. My default is that every write waits for a person to confirm, and every step is audited.

Related services

Fractional CTO
Technology leadership without a full-time executive hire.
Read more →
Cloud and infrastructure
AWS, Cloudflare and VPS architecture, migrations and cost control.
Read more →
Platform and product engineering
Multi-tenant SaaS, web and mobile products, built to scale from the first commit.
Read more →
Next step

Start with AI and agentic systems.

Book a consultation and say what you need. The first call is about your situation, not a sales pitch.

Book a call →[email protected]