IT Director in Dubai. Site networks, industrial control systems, physical security and the Microsoft estate above them.
Systems I scoped, procured and commissioned. The scale is the part that is hard to fake.
Ninety cameras across thirteen construction sites, monitored live from one room by one operator.
Every site connects over 4G behind carrier-grade NAT, so no inbound port forwarding is possible. Built on outbound P2P instead, with live sub-streams tuned for the worst case mobile link and a video wall laid out by risk rather than by site.
Read the case studyWeeks of random plant stoppages traced to one loose contact on an INTERBUS ring nobody had a current map of.
Every event carried the same signature: one gateway error, then ten alarms that were all consequences of it. Rebuilt the ring topology from three mixed drawing revisions, then induced a known fault on site to prove the failure was a module rather than the cable between cabinets.
Read the case studyRecovery and stabilisation of an Ammann AS2000 asphalt plant control system running on SunOS 5.8.
Rebuilt a corrupted System V startup chain from init through rc2.d to S99uag, restored the application environment, and resolved shared memory, permission and watchdog faults behind a respawn loop. No vendor support, no rollback, and every restart a production decision.
Read the case studyComplete IT infrastructure for a highway construction site, built from nothing with no internet and no cloud.
Multi-VLAN network with router-on-a-stick routing, a Grandstream IP PBX registered across VLANs without Zero Config, a 24 TB NAS with role-based access for twenty-four users, and ONVIF surveillance. Designed to work with the outside link down, then connected.
Read the case studyMicrosoft retirements I am planning around. Every one has a fixed date and a client side consequence somebody will have to absorb.
| Date | What changes | Status |
|---|---|---|
| 31 Mar 2026 | Basic authentication retired in Exchange Online Every client, script and service account still sending a username and password stops working. If something broke quietly in April, this is usually why. | Passed |
| 27 Jun 2026 | Secure Boot 2011 certificates begin expiring Staged through October. Devices without the 2023 CAs stop trusting newly signed boot components. Existing installs keep booting, but the fleet stops accepting new updates. | In progress |
| 1 Oct 2026 | Exchange Web Services disabled by default Tenants that did not set an AppID allow list get EWS switched off. Migration tools, backup products, room booking and CRM connectors are the usual casualties. | Act now |
| 13 Oct 2026 | Windows 10 consumer ESU ends Enterprise customers can renew yearly to October 2028. Anything outside that programme stops receiving security updates entirely. | Act now |
| 1 Apr 2027 | Exchange Web Services permanently retired No allow list, no re-enabling, no exceptions. Anything not moved to Microsoft Graph by this date stops working and cannot be turned back on. | Plan now |
| Oct 2028 | Windows 10 enterprise ESU ends for good The final renewable year. This is the real deadline for fleets that could not make the hardware jump in time. | Plan now |
Three things I built and still maintain. One repairs an estate, one runs a marketplace, one saves an afternoon.
Repairs the client side after an on-premises Exchange to Microsoft 365 cutover. Password loops, stale Autodiscover redirects, dead MAPI profiles.
Runs as a standard user with no admin rights. Every change previews first, backs up the registry, and reads the value back to confirm it took.
A Berlin neighbourhood help marketplace. Neighbours post what they need, neighbours nearby offer to do it, both sides agree the price.
Duplicates a site locally with its assets and link structure intact, for archiving and offline reference.
Usually after something has already gone wrong, which is the part of the job I like.
Networks built where there is no internet, control systems older than the staff, and cameras across sites you cannot drive to.
Identity, mail and endpoints, and the migrations that move them without a week of tickets afterwards.
Owning the estate end to end: vendors, budget, security posture, and the people who answer the phone.
Notes from the estate, mostly written the week after something broke.
Disabled by default on 1 October 2026, gone entirely by April 2027. The allow list buys you six months, not a solution.
Fifteen year old certificates, staged expiry from June, and a failure mode that does not announce itself at boot.
Three renewable years is enough time to migrate properly and exactly enough time to keep postponing it.
Your users are already building them. The permission model they inherit is the one you set years ago and forgot.
Six lookup methods, first answer wins. After a migration the second one is usually a server that no longer exists.
Three caches hold the previous identity, and clearing the obvious one fixes it for about a day.
I am Ahmet Mustafa Yılmaz, an IT Director based in Dubai. The estate I run goes from the plant floor to the cloud: industrial control systems on hardware older than some of the staff, site networks built where there is no internet, camera systems across a dozen construction sites, and the Microsoft tenant everything else signs into.
Most of what I write about is what broke and how I found out why. The tools I publish exist because a problem repeated often enough that fixing it by hand stopped being defensible. Outside the estate I build KiezHelfer, a neighbourhood help marketplace in Berlin, which has taught me more about product decisions than a decade of infrastructure did.
Migration work, an estate that has outgrown its last design, or a question about the toolkit. All of it is welcome.