Service · pilot-first, measured output
AI & automation that actually ships.
Practical AI features and workflow automation for businesses that want time back and better output — not a chatbot stuck in the corner of the website. Every engagement starts with a pilot that proves the AI is reliable enough for your specific input before we scale.
What I do
Five concrete things.
AI product features inside your app
Reasoning features powered by the Claude API with structured JSON output, so the output is reliable enough to drive business logic rather than just a chatbot UI. Example: NoonSpy's 'Verdict' feature — paste a product URL, get a Sell / Risky / Skip decision + five scores + a written next action.
Product content assistance
Bilingual listing writers (English → Arabic + English), description generators, image-pack generators via Gemini. Human review built in — the AI drafts, you approve before publish.
Workflow automation
Make.com / n8n / custom Python scripts that remove a repeated hour from your week — invoice reminders, inventory syncs, lead triage, calendar routing. Scoped to one workflow per pilot, measured by time saved.
Enquiry organisation + draft responses
Email + WhatsApp enquiry triage: categorise, extract the key fields, draft a first-response. You review and send — AI never ships a message under your name without your eyes on it.
Integrations between tools you already use
Stripe → Google Sheet. WhatsApp Business → CRM. Shopify → Zoho. Only integrations where both sides have an official API and your account can authorise the connection. No hacky scraping of someone else's platform.
How I run AI projects
Four honest rules.
Pilot first, scale after
Every AI engagement starts with a one-workflow or one-feature pilot (typically 1-2 weeks, AED 2,000-5,000). We prove the output quality against real data from your business before scaling. If the pilot shows the AI isn't reliable enough yet, I say so and we stop — not everything that's AI-doable is AI-worthwhile today.
Human review is default
Any output that reaches a customer, invoice, or external system goes through your review by default. Autonomous AI (acts without review) is only used after we've measured enough runs to trust the output. 'Agent that does everything for you' is often a project you should not pay for yet.
You own the API account
Claude / Gemini / OpenAI keys live in your account, not mine. You see the monthly bill directly, you can revoke access any time. I set up the account, you fund it. Typical cost: $20-200/month for a small-business workflow, depending on volume.
Failure handling in the spec
What happens when the AI returns garbage? When the API is down? When the input is in a language the model struggles with? Every pilot answers these upfront — fallback to human, retry logic, logged errors you can see.
Real examples
AI I've actually shipped.
These are my own products — not client engagements. I show them as evidence that the engineering is real, not borrowed from a template. For client AI projects I follow the same patterns on your own stack.
NoonSpy (my own SaaS)
AI Product Verdict feature — paste a Noon marketplace URL, get a Sell/Risky/Skip decision in ~10 seconds, grounded in live marketplace data rather than general-purpose chat. Built on Claude API with structured JSON output; paid subscribers across UAE and Saudi.
Owned product — I built it to solve a problem on my own Noon store, then packaged it.
Visit →ArabicListingWriter (my upcoming product)
Converts English product details into bilingual Arabic + English product listings in 10 seconds for Noon / Amazon UAE / Shopify sellers. Native-Arabic quality via carefully-tuned prompts + validation. Waitlist open, launching November 2026.
Owned product in beta.
See the page →Hanooon image-pack generator
Gemini Nano Banana Pro pipeline for e-commerce product imagery — takes a product photo and generates lifestyle shots, catalogue tiles, social-ready crops. Used on my own metal wall-art brand (not sold as a client service yet).
Owned pipeline. Available as a bespoke build for other brands on request.
Visit →FAQ
Honest answers.
What AI tools do you use?
Claude API (Anthropic) for reasoning, structured JSON output and content generation. Gemini (Google) for image generation and multi-modal work. OpenAI / GPT when a client already has credits there or has a specific feature compatibility requirement. Make.com, n8n and custom Python for workflow automation. All work runs on your own API keys — you fund the usage directly.
How much does an AI project cost?
A one-workflow pilot (1-2 weeks of work) is typically AED 2,000-5,000. A production AI feature inside an existing app is AED 5,000-15,000 depending on complexity and the amount of guardrails needed. Monthly running costs (API bills + any hosting) are typically $20-200/month on top, paid directly by you to Claude / Gemini / your hosting.
Can you replace my customer-service team with AI?
No, and I'd argue against anyone who says yes. AI can draft first-responses, triage enquiries, and route them to the right person — that's a real workflow improvement worth doing. Fully autonomous customer service (AI sends replies with no human review) is a reputational risk I don't recommend for most small businesses in 2026.
What if the AI output isn't good enough?
That's exactly what the pilot phase is for. We test the output against real inputs from your business and measure accuracy. If it doesn't clear the bar you set, we stop and refund the pilot deposit minus the hours already worked. Not every AI project should ship.
Do you build 'agent' systems?
Technically yes — Claude's tool use, structured outputs and workflow orchestration are things I've built with. But for small businesses, agent systems that act autonomously are usually a worse ROI than well-scoped single-task AI features with human review. I'd rather sell you something that definitely helps than something that sounds impressive.
What data do you need to get started?
For a pilot: 20-50 real examples of the input the AI will see (customer messages, product descriptions, orders — whatever the workflow processes). Private data stays in your systems; API calls go directly from your account to Claude/Gemini. For some workflows a small test dataset is enough; for others we need a week of real traffic.
Project brief
Send me the brief.
Fill the form, then pick WhatsApp or email — the form composes a structured brief and opens your chosen channel with the message pre-filled. Review before you send. Reply within 24 hours.