Product engineer building trustworthy AI automation and data systems.
Founder of Kelyvaro — from first prototype to production.
Karlsruhe · Türkçe / Deutsch / English
I turn operational problems into products: finding and normalising data from unreliable external systems, generating documents without inventing facts, and shipping the result with clear consent and failure boundaries.
Current focus: Kelyvaro, a live product that helps people in Turkey apply directly to German employers. It matches openings from employer career sites, prepares a German CV and cover letter for each role, and sends only after the user approves the application.
| Project | Problem | What I own | Proof |
|---|---|---|---|
| Kelyvaro | A careful German application takes too long and is easy to get wrong | Product and engineering: matching, fact-checked generation, consent, data model and design system | Live product · case study |
| ats-boards | German employers publish jobs through incompatible ATS providers | A dependency-free TypeScript library and CLI that discovers and normalises public boards from 12 providers | Source · npm |
| keepwatching | Short-form format advice often presents guesses as measured facts | A deterministic renderer and evidence model that keeps hypotheses, samples and content effects separate | Source · live gallery |
- Facts stay outside the model. Generated CV claims are checked against the user's profile; prompt changes have regression tests.
- Public integrations fail explicitly. ATS results distinguish an empty board from access, rate-limit and format failures.
- Consent is part of the architecture. Applications require user approval; analytics are consent-gated; database tables use row-level security.
- Evidence keeps its sample size.
keepwatchingpublishes untested formats as untested and runs its date reader in four time zones.
TypeScript · Next.js · React · Node.js · Postgres / Supabase ·
Playwright · Tailwind CSS · Vercel · Remotion