About
I Build the System, and I Write Down Why
My name is Muhammet Şafak, and I am a software engineer. I have been building software since 2008, and today I work as a Staff Engineer on the R&D team of a health tech company. Thinking about and designing high-traffic architectures is the part I enjoy most, and investigating how systems actually behave through experiments and measurement is part of the job. My current direction is building and operating AI models as components of production systems. I am loyal to the problem, not to the language. What I know, what I learn and the real-world cases I run into, I share in several places — this site first of all.
This page tries to explain both who I am and why this site exists.
Staff Engineer Istanbul, Türkiye
What I decide by
I have been at this a long time and plenty has changed; the ones below have not. Each one has a cost, and I have written that down too.
- 01 I choose what can be changed tomorrow
- The thing that finishes fastest today is rarely the thing still standing a year later. When the two conflict I take the second one; the price is a slower first delivery.
- 02 I write down the reasoning
- Making the call is half the work; writing down why it was made is the other half. An unrecorded reason turns into code nobody can defend six months later.
- 03 I name the cost
- Every choice comes with a bill. A recommendation that hides its bill is not advice, it is sales copy. Nothing here is a cure-all.
- 04 I am loyal to the problem, not the language
- PHP, Go, Python, JavaScript and Rust all stay on the table; the problem decides which one opens, not my habit.
- 05 I say so when I do not know
- If I have a limit on a topic, I do not hide it. Seniority is not knowing every answer; it is drawing the line in the right place.
Where I go deep
The subjects I have read into, measured, and shipped production work with. On anything outside this list, I will tell you I do not know.
-
Systems architecture
In high-traffic systems, scalability and reliability are design decisions, not features added later; the reasoning and the cost of each decision are written down as part of the design.
-
Developer tooling and CI
Tools that do one thing well, and the pipeline that runs them. A green CI is not enough; what it proves has to be written down too.
-
Event-driven and messaging
The event itself carries the contract between services; queues, retries and idempotency are the bill that comes with it.
-
AI systems engineering
I treat the model as one component of the software system that carries it. My principle: the machine works at the language layer, not the decision layer. Code and people make the decision; the real work is knowing what the model does not know.
-
Automation and data processing
Every manual repetition is an error budget. Collection, transformation and scheduled jobs; what separates a small script from a durable pipeline is how it handles failure.
How did I get here?
I started freelancing in 2008. For fourteen years I delivered corporate sites, e-commerce systems and integrations. That is where I learned the commercial side of the work, what a delivery date actually means, and that no system survives without an owner.
From 2012 I was also on the core team of a vertical search engine: an index of tens of millions of documents, a crawler measured in millions of URLs an hour, and query latency dragged from seconds down to milliseconds. That is where I first saw what “scale” means, bill included.
Since 2022 I have been full-time at a health tech company — first as a Senior Software Developer, and since 2025 as a Staff Engineer. The job is no longer only writing code: designing the microservice and event-driven architecture of a high-traffic platform, acting as the technical compass on critical design calls, and putting product strategy and technology vision on the same table. Over the same period I mentored small teams on TDD and DDD.
I did not learn software at school — my degree is in history. That started as a gap and turned into a habit: read everything at its source, and note down what you read with the date on it. That habit explains a fair amount about what this site looks like.
Company names, dates and the skill list live on the résumé; this page is about the work itself.
See the résuméAbout this site
Built with Astro, Tailwind CSS and MDX, published on Cloudflare. The content does not live in a database but inside the git repository, as markdown.
No account to open, no member dashboard and no ads — none of them planned either. Just Ask has a question form, and comments run through giscus with a GitHub sign-in.
Tan, the beaver on covers and share cards, has a dedicated page explaining who Tan is and why a beaver was chosen.
For anything else, such as an offer or plain curiosity, write by email — there is no separate contact form.
Write to meThe archive
Posts dated before 2026 were first published in their own years, on a previous version of this site or on another platform. In 2026 I went through all of them and brought them together here, much as Engineering moved over from sade.dev. The year in each post’s box is the year of first publication. In that review I made edits to headings, links and formatting, and added sources and summaries to some posts. The texts keep the knowledge of the year they were written in; where something has aged or turned out wrong, I mark it with a dated note inside the post. The English editions were added in 2026.
What AI does to these posts
AI is not here to write the text; it is here to anonymise what I wrote and clean up the language. Every post I publish today goes through three steps.
- 01 Draft
- I pick the topic, and I write the text from cases I actually hit in production. The draft cannot be published as it stands: whatever makes the customer, the system or the data recognisable is still in it.
- 02 Scrubbing
- AI anonymises the text: it generalises anything recognisable without breaking the example. Typos and clumsy sentences are fixed in the same step.
- 03 Final review
- Before it goes live I read the text line by line, fix whatever needs fixing and give it its final shape.
This is where AI does not reach: the topic, the claim, the real examples and the measurements are mine. The machine works on the language layer, not the decision layer.
What lives where
This site is not a showcase, it is a ledger. It is not one pile either: each section answers a different question and none of them does another’s job. Your question decides where you should go.
-
Blog
How did I build this?
A hands-on development journal, dated from 2014 to today. Every post stays inside the knowledge of the year it was written; I do not go back and correct the past with what I know now.
-
Labs
What did I build, how, and where did it end up?
The development centre. Each record follows how a piece of work was built: most became products and moved to the portfolio, leaving their record here; on some development stopped, and the record says so.
-
Research
Which question did I chase, and what did I find?
The research centre. Measurements, analyses, surveys and reviews — each record states its method, its sources, and how far I trust the result.
-
Portfolio
What shipped?
Work that actually went live: what it does, who it is for, and whether it is still alive. A record, not a pitch.
Engineering
The line is drawn by mode, not by person
Everything written on the blog is dated and first-person — “I did this, on this date, this way.” The Engineering section speaks by topic — “this is how it is done” — with its costs and limits. The same topic can live in both; one carries the finding, the other the principle. When something on the blog touches system-level depth, I do not write that depth there — I link to Engineering.
Go to Engineering