Systems engineer. Seven years on Turkey's GİB e-Transformation infrastructure — e-Invoice, e-Archive, e-Ledger, fiscal-seal signing.

I build multi-tenant regtech systems: the kind that process ~20,000 e-invoices a day, sign them, and submit them to a tax authority where a malformed field has legal consequences. Most of my building now is AI-native — agentic systems, the Claude Agent SDK, my own skill/agent/MCP stack — on top of a domain I spent ten years learning by hand. That depth is what lets me tell when the machine is wrong.

Stack

TypeScript/Node · NATS/JetStream · Temporal · RabbitMQ · ClickHouse · PostgreSQL · Docker · K3s · PKCS#11 / XAdES · Claude Agent SDK / MCP

Open source

• json2ubl-ts — UBL-TR 2.1 generation: 53 Schematron rules, 12 invoice profiles, 20+ invoice types, a reactive InvoiceSession.
• ubl2json-ts — the reverse: UBL-TR into structured data.
• yabbistir — a keyboard-first clipboard history for macOS: it opens right under the text caret, finds anything you copied as you type, and pastes with ↵. Categories, a Stack you can reorder, transforms, and JavaScript extensions from a signed store. Swift 6.2 + SwiftUI; notarized DMG or Homebrew.
• mystat — a calm system monitor for the macOS menu bar: it tells you in one sentence what's going on, and who's responsible when something isn't right. Swift 6.2 + SwiftUI; notarized DMG or Homebrew.

Now

Building agentic infrastructure that pairs deep GİB regulatory knowledge with autonomous agents. Open to partnerships, advisory, and good technical conversations.

Links — LinkedIn · npm

Most of my work lives in private org repositories, so the public graph shows only part of it.