I connect AI to the system you already run.

Who is Savaş Doğan and what does he do?

Savaş Doğan is a freelance software architect and AI integration consultant based in Türkiye. He has been building software since 1992. He does three things: adds LLM-based capability to existing enterprise software, builds multi-tenant products with Angular and Node.js, and works as a fractional CTO on technology and AI adoption decisions. He works remotely, in English and Turkish.

Plenty of people do one of those three. The practical value of finding them in one person is this: the person who designs the architecture, writes the code and defends the budget is the same person. Nothing is lost in translation.

First line of code

1992

In the industry

33+ years

Executive

CIO / VP · international game studio

Founded

BTOS · 6 years

Working languages

English, Turkish

Timezone

UTC+3 · overlaps EU hours, partial US East

What can you hire me for?

Three services: AI integration, product and software development, and technical consulting. Each can run as a fixed-scope project, a monthly retainer, or a long-term remote contract with a fixed number of days per week. Every service page states scope, duration and engagement model explicitly.

What can I actually prove?

Three pieces of work are documented on this site in technical detail: a multi-tenant CRM shipped to production with over 4,000 automated tests, my own benchmark series measuring what open-weight models can and cannot do on 48 GB of unified memory, and a local-first desktop engine where data never leaves the device. All three are written with numbers, not adjectives.

Why does experience matter in AI integration?

Wiring an LLM to an API call takes a few hours. The hard part is everything around it: which data reaches the model, where a wrong answer lands, how cost is capped, and who is allowed to see what. None of those are AI questions. They are architecture questions, and they are the ones that decide whether a pilot ever reaches production.

• Data boundaries. In a multi-tenant system every query must be scoped by tenant. If that discipline is not extended to the LLM context, one customer’s data ends up in another customer’s answer.
• Cost of a wrong answer. If a flow produces prices, stock levels or legal text, every output must trace back to a verifiable source. If it cannot, that flow does not go to an LLM.
• Cost ceiling. Token cost grows linearly with usage. Unless you decide up front which step runs on a local model and which on a hosted one, the project clears the pilot and fails the rollout.
• An exit path. Providers change prices and retire models. Without a provider-agnostic interface layer, that change means a rewrite.

How an engagement runs

Four steps: discovery call, written scope and proposal, development in two-week slices, handover. The discovery call takes 30–45 minutes and is free. No development starts before the scope is approved in writing.

1. 01

   Discovery

   We talk about the problem you are solving, not the feature you want. 30–45 minutes, free. You leave with a written answer to three questions: can it be done, should it be done, and roughly how long — whether or not the work comes to me.

2. 02

   Scope and proposal

   Scope, deliverables, schedule and budget in one document. What is explicitly out of scope is written down too — that is where the disagreement always turns out to be.

3. 03

   Development

   Two-week slices. Each slice ends with a working demo and a short note on what is done. Your access to the repository is continuous; you never have to ask for a status update.

4. 04

   Handover

   Deployment, an operations document, and a codebase your own team can take over. Post-handover support is written into the contract rather than left vague.

Frequently asked questions

How long does an AI integration project take?

A typical AI integration project takes 4 to 12 weeks. A retrieval-augmented search over internal documents sits near the lower bound; agent workflows that connect several systems sit near the upper bound. What drives the duration is not model choice but how clearly data access and authorisation rules are defined.

Can you work with a team in another timezone?

Yes. I am at UTC+3, which overlaps a full working day with Western and Central Europe, the UK, the Gulf and India, and the morning of the US East Coast. I work asynchronously by default: written scope, written slice notes, and a repository you can read at any hour.

Do you take long-term contracts, or only projects?

Both. Fixed-scope projects are my preference when the scope is clear. For a team that needs a senior engineer or a technical lead embedded for 6–12 months, I take long-term remote contracts with a fixed number of days per week. Monthly retainers cover ongoing consulting and maintenance.

Will you take over an existing codebase?

Yes, and a meaningful part of my work is exactly that. Taking over an undocumented codebase is a different skill from writing one: first map what actually works, then change the riskiest part without breaking it. I start with an audit and report findings at file and line level before touching anything.

How do you handle contracts and invoicing for clients abroad?

I work under an English-language contract and invoice as a registered self-employed professional in Türkiye. Payment is by bank transfer or an international payment provider. An NDA can be signed before the discovery call if your process requires it.

Last updated: September 18, 2026