mert@msa:~$ id --operator
Mert Sefa Akgün
mert@msa:~$ cat mandate.txt
Lead software engineer · AI systems
Real-time AI, voice, memory, networking, and embedded software. I take the fragile path, name every failure, and turn it into an observable system I can trust under pressure.
Three systems. Three proof surfaces. I publish the constraint, the boundary, and the recovery path—not a list of technologies arranged for applause.
I architect the live path of production voice agents—from audio ingress and interruption handling to tool execution, memory, recovery, and stage-level telemetry.
scope: Production architecture, published without customer details
Latency ownership · bounded queues · cancellation · degraded modes · OpenTelemetry
inspect proof ->I built a memory policy that keeps source, trust, sensitivity, allowed use, expiry, and auditability attached to retrieval instead of surrendering them to an embedding.
scope: Agent infrastructure with explicit permission boundaries
Typed policy · conservative defaults · side-effect gates · lifecycle tests
inspect proof ->I designed a field RTLS that combines UWB ranging, Bluetooth provisioning, filtering, and GPS/time sync under access-control hardware constraints.
scope: Embedded/RF systems under real deployment constraints
TDoA/TWR · EKF · calibration tooling · device integration · field debugging
inspect proof ->I write to preserve decisions, expose failure modes, and leave behind something sharper than a retrospective. Press / to search; Esc clears the field.
I treat voice-agent latency as a chain of accountable decisions—not a flattering average on a dashboard.
I refuse to let embeddings erase source, trust, sensitivity, permission, expiry, or the right to say no.
A compact C++ failure mode: global initialization crosses translation units, and startup becomes link-order roulette.
Stack or heap is not a trivia question. It is a promise about ownership, lifetime, movement, and cleanup.
The radio can measure distance. The product still has to survive geometry, calibration, provisioning, clocks, enclosures, and the field.
Before the primitive, I want the adversary, the asset, the channel, and the exact cost of failure on the table.
A small algorithm file on universal integer coding, representation choices, and the value of keeping primitives testable.
Editors, shells, agents, and automation are not accessories. They are the control plane around how I think and ship.
Autonomy without state, approvals, interruption, and recovery is not power. It is a process I cannot trust.
I started where mistakes become physical: firmware, radios, constrained devices, and networks with hard timing. I carried that discipline into real-time AI. The stack changed; the standard did not. I want explicit owners, bounded failure, measurable recovery, and evidence that survives contact with production.
The happy path is cheap. I design the cancellation, pressure, and recovery path first.
I do not confuse retrieved context, instinct, or confidence with permission to act.
Per-turn traces, queue age, and percentiles tell the truth an average will not.
I bound side effects, expose state, and make rollback the least dramatic part of the system.
mert@msa:~$ ./open-direct-line --only hard-problems
If it touches latency, concurrency, memory, observability, embedded systems, or an agent that must recover cleanly, I want the hard version.