::03 /proc/self/status

Specialization follows consequence, not layer.

The work started where mistakes become physical: firmware, radios, constrained devices, and networks with hard timing. That discipline now governs real-time AI. The stack changed; the standard did not: explicit owners, bounded failure, measurable recovery, and evidence that survives contact with production.

Name
Mert Sefa AKGUN
Role
Lead software engineer / AI systems
Range
firmware → kernel → realtime agent
Toolchain
Rust · Python · TypeScript · C/C++
Standard
explicit owners · typed states · measurable recovery
Current
voice · memory · observability · agent control
01Own the failure path

The happy path is cheap. Cancellation, pressure, and recovery are designed first.

02Move with evidence

Retrieved context, instinct, and confidence are evidence—not permission to act.

03Make pressure visible

Per-turn traces, queue age, and percentiles tell the truth an average will not.

04Keep exits boring

Side effects stay bounded, state stays visible, and rollback remains the least dramatic part of the system.

::03.1 /home/msa/story

I build to understand the whole system.

I am a lead software engineer based in Istanbul. Today I work on real-time multimodal AI platforms: voice, tools, memory, retrieval, persistence, and the operational paths that keep long-lived sessions dependable.

Before that work moved up the stack, it moved through it—production infrastructure, embedded and RTOS firmware, access-control protocols, UWB localization, Linux networking, TSN, eBPF, and XDP. I still see those areas as one continuum. Each layer changes the constraints, not the obligation to understand what the system is doing.

That is why I keep building outside the immediate job: an operating system, a programming language, ray tracers, a WASM/WebGPU game, control surfaces, small electronics, and the infrastructure behind the tools I use. Building from scratch turns abstractions back into mechanisms.

::03.2 /home/msa/practice

What I care about away from the editor.

Engineering is the main practice, not the whole identity. Sound, physical making, and time outdoors keep curiosity attached to the senses and to the real conditions a clean abstraction can hide.

01 / sound

Music and sound

I am an audiophile and make electronic music. Music production and EDM sound design—rumble kicks, trance synths, dynamics, and texture—are a parallel craft where timing, restraint, and tiny changes are immediately audible.

listening · synthesis · dynamics · texture
02 / physical

Maker practice

I like ideas that can leave the screen. Electronics, hardware/software prototypes, 3D modeling, firmware, home automation, and tactile control surfaces give me a direct loop between an interface and physical behavior.

electronics · controllers · fabrication · automation
03 / outside

Outside

I cycle, go hiking and camping, climb, and snowboard. I am drawn to activities where terrain, weather, momentum, risk, and the limits of the body stay present—conditions that cannot be abstracted away.

cycling · hiking · climbing · snowboarding
::03.3 /home/msa/next

The common thread is deliberate practice.

Whether the medium is a low-latency agent, a packet path, a sound, or a route through the mountains, I am drawn to work where feedback is real, constraints matter, and quality is felt before it is explained.