Skip to content
Salim Kızılay Türkçe — view this page in Turkish

I learn your process on site, then write software that fits it

Eighteen years inside manufacturing and trade — production planning, quality, foreign trade, business development. Today I analyse those same processes and build the AI integration and custom software they need. I do not start from an off-the-shelf product; I start by watching how the work actually runs.

How I work

01

I start on the floor

I talk to the person who runs the process every day. What stayed with me from the field study for my thesis — 45 company representatives across five industrial zones — was not a conclusion but this method.

02

The diagnosis is written down

What should be automated matters as much as what should not. We settle the scope and the order before any code is written.

03

I take it to production

A working prototype is not enough. Security and data-protection compliance are not bolted on later; they are part of the architecture from day one.

Engagements

See all

I map how your company actually works today, on site. Software recommendations come last.

2–3 weeks

  • Field interviews with the people who use the system, not only management
  • A map of the current workflow: where data is entered, re-entered and lost
  • Digital maturity assessment across speed, flexibility, competence and responsiveness
  • A prioritised intervention list: what each automation would actually be worth
  • A do-not-do list — where spending money would be wasted
  • Software development — no code is written at this stage
  • Software or hardware resale; I sell no products, so the advice stays independent
  • Staff assessment or HR consulting

I add one AI capability to your existing software and take it all the way to production. Two weeks.

2 weeks

  • One concrete capability integrated end to end and running in production
  • A connection into your current system — no migration, no platform change
  • Cost measurement: what each call costs and what the monthly bill will look like
  • A designed failure path for wrong answers, including a human approval step
  • Source code and documentation are yours; you are not locked to me
  • Training a model from scratch — I use existing models
  • More than one capability at a time; a sprint is deliberately narrow
  • Rewriting your existing system
  • Long-term maintenance, which is a separate engagement

I turn the need found in the diagnosis into software built around your workflow, not a package you adapt to.

Monthly, three-month minimum

  • An application designed around how your company actually works
  • A working release every month — regular small steps, not one big delivery
  • Production setup: deploy, backups, monitoring and a rollback plan
  • Security and data-protection review of authentication, authorisation and data flow
  • Handover training and documentation for the people who will use it
  • Fixed-price, fixed-scope projects — scope is agreed monthly
  • 24/7 on-call support
  • Hardware, networking or system administration
  • Taking over software written by another team

Delivered work

Consolidating a training centre that ran on four separate programs

Driving school, professional competence, psychotechnical assessment and dangerous-goods training were tracked separately. The same candidate was registered again in every module, and payment figures differed depending on which panel you opened.

Driver training and vocational certification · 2026

Training delivered

10 August 2026, Coldwell Banker Siam Gayrimenkul office. A 23-second clip with no sound: the broker team around the meeting table, laptops open in front of them, following the session.

I ran a two-hour AI training for the broker team at Coldwell Banker Siam Gayrimenkul. The question was simple: without buying a new product, how much of the work can run on its own using the Google tools the company already pays for? Nobody sat through slides — everyone opened a laptop and set up all four tools in their own account.

Sheets + =AI()

The listing headline, the description, the Instagram post and the WhatsApp message for a new property all come out of a single row.

Forms + Apps Script

A buyer request submitted through the form is matched against the portfolio, and the three best-fitting properties land in the broker's inbox.

NotebookLM

Answers questions on title-deed fees, mortgages and zoning strictly from the company's own documents, showing the source under every sentence.

AI Studio

An objection-handling drill and a post-meeting summary app — built by describing them, without writing code.

It closed on four rules: AI writes, the broker signs; no number is used without a source; client data is not pasted around carelessly; the hour you save goes back to the client.

There was a trap on the final slide. For sixteen slides the heading had misspelled the Turkish word for “control” and nobody had caught it — and the deck itself had been written by an AI. That was the lesson: the same slip can land in a listing, in a contract summary, in an email to a client. There it costs more than a word.

Who I am

Salim Kızılay

I started on the technical side — marine electronics and communications, then serving as an electrical officer at sea. Then industry: production planning at Avery Dennison, quality at Arçelik. I ran my own wholesale and retail business for six years. Today I am a shareholder at Siam Group, running global operations for five pet-care brands.

Across all of it I kept seeing the same thing: companies with good products stall because they cannot see their own processes.

About CV Contact