Oğuz Samsa

I'm a full-stack engineer who works in Go on the backend and TypeScript with React and Next.js on the front end. I came to software through a bootcamp after a first career in mechanical engineering, and I've spent nearly two years since on one thing almost exclusively: building AI-powered products end to end.

Most of that was a single product: an AI video-generation platform for mobile game marketing. I wrote its first commit at Fill-Labs, the agency that started it, and when the client took the project in-house they hired me directly to keep building it. Across those 13 months I was one of its two engineers. On the backend I built the asynchronous generation pipelines in Go — queue, worker, consumer — with transactional enqueueing, retry handling, timeout recovery, and per-request usage metering. The piece I'd point to first is the integration layer: five third-party AI providers (OpenAI, Google Gemini and Veo, fal.ai, ElevenLabs, AssemblyAI) behind a single Go interface, plus a routing layer that brought onboarding a new video model down to an isolated, five-step change. On the front end I was the primary author of the Next.js 15 application, including a timeline-based video editor and live generation status over Server-Sent Events.

The work I learned most from was an incident where video exports silently stalled in production and threw no errors anywhere — the Lambda logs showed every render succeeding, so I spent the first stretch looking in the wrong place. The render records in the database turned up two causes instead: a global render-slot counter that leaked slots on certain failure paths until the pool was exhausted, and a race where the front end's status polling could start the same job twice, so a 16-clip export dispatched 32 renders. A janitor goroutine to reclaim orphaned slots and a per-job mutex on startup closed the whole failure class. More recently I built a small Laravel and PostgreSQL token-wallet service to work through billing correctness properly: webhook deduplication, row-level locking around balance changes, an append-only ledger with database-level invariants, and a concurrency test that asserts no double-spend under parallel requests.

What I'm looking for is a mid-level full-stack role on a small team where I own features end to end rather than picking up tickets. I'm most useful where a product meets messy third-party APIs — auth flows, rate limits, partial failures, vendor specs that turn out to be wrong — and where it helps to hold both the backend and the interface in your head at once. I use AI coding agents daily and treat their output as unproven until tested. I'm based in Türkiye (UTC+3) and work remotely, with full overlap across European time zones and partial overlap into US mornings.