← KAIRA PRIME kairaprime.com/cv

Ufuk Kazan

Software Engineer — backend, systems & AI-tooling

Belgium (remote) Dutch (native) · English hello@kairaprime.com kairaprime.com

Profile

I'm a self-taught software engineer in Belgium. I didn't come through a CS degree — I learned by building, and by holding my own work to a hard bar: real tests, real architecture, systems that stay correct when a worker crashes and retries.

I work AI-orchestrated: I own the specification, the architecture and the tradeoffs, and I direct AI (Claude Code) to implement, under a strict spec → build → test → review loop. That lets me ship production-grade backends fast — my commerce platform Devicely went from zero to 356 passing tests, solo, in about three weeks. I care most about the work where correctness and judgment matter: backends, systems, and AI tooling.

Selected work

DevicelyBackend / platform · solo · ~3 weeks

Event-sourced commerce platform for reselling refurbished laptops across Europe's marketplaces (Refurbed, BackMarket, Bol, Cdiscount). Commands flow through an outbox with retry + reconciliation, so a crashed-and-retried worker can't double-list or double-charge. Local on-device vision QC gates every product photo; a profit engine encodes Belgian tax rules.

PHP 8.3Event sourcingLocal VLM356 tests / 2,551 assertions
Kaira OSSystems · Rust + Python

An AI-native OS shell with a behaviourally-verified Rust authority core: a 7-level permission engine (L0–L6) that actually refuses a reboot and waits for consent, a cross-process kill-switch, and a local-first daemon that opens no TCP ports.

Rust (tokio, 11 crates)~92k Rust + ~85k Python~1,300 + ~4,900 tests
Soulmask build optimizerFull-stack + custom tooling

A game build-optimizer backed by real extracted data. When no tool could read the game's Unreal Engine .uasset files, I wrote a custom .NET 8 extractor on UAssetAPI to pull them out.

.NET 8 / UAssetAPIFastAPI · 906 testsReact + TypeScript

A self-hosted pipeline that ingests and ranks data through a versioned, auditable scoring formula — built to be operated and audited, not trusted blindly.

FastAPI · SQLAlchemy21-table schema · 65 migrations~3,000 tests
Hardware test stationSystems / automation

Boots and tests up to 36 refurbished laptops in parallel over USB/PXE, each self-reporting to one dashboard with a deterministic grade and an AI-drafted triage note. Written in pure standard-library Python — zero external dependencies, offline-capable.

Python (stdlib only)PXE / USB bootUsed on-site
RigDeckAI tooling · daily driver

A desktop cockpit that drives my remote build rig over SSH/Tailscale, with a "brain" that embeds every repo on the rig and answers semantic code queries.

Python · GTKLocal embeddingsSemantic code search

How I work & what I use

Languages

PHP · Python · Rust · TypeScript / JavaScript · C# / .NET

Domains

Event sourcing & distributed-systems patterns · permission / authority engines · AI tooling (MCP, agent orchestration, local LLMs) · full-stack web · data pipelines · hardware automation

Practice

Spec-driven & test-first. AI-orchestrated delivery: I hold the architecture and the judgment; AI does the typing, under review.

Building now

Kaira Prime — an independent studio shipping small, well-kept software products, built in the open.

Looking for

Remote backend / platform / AI-tooling roles (contract or salaried), ideally on teams that value correctness, ownership, and an AI-native way of working. Based in Belgium; Dutch native, English fluent; comfortable across EU timezones.

Ufuk Kazan · hello@kairaprime.com Full case studies at kairaprime.com