SDET at RECETOX, Masaryk University. I build web applications and the tests that prove they work. A change is done when a test says so, not when it looks right.
AI writes code faster than people do. It writes bugs faster too. So the model drafts, and the tests decide.
I run open-weight models locally on an Intel Arc B70: as coding agents, for analysis, and to find where they break.
Work with: Vue, Nuxt, TypeScript, Python, Django, some Symfony, PostgreSQL, shell scripting · Playwright, Cypress, CI/CD · Claude Code, MCP servers, local models as coding agents · Debian, because it is stable and does what it did yesterday.
Learning: React.
Off the keyboard: electric guitar.
- cypress-mcp: MCP server. Gives an AI assistant the Cypress specs and the results of the last run.
- po-generator: crawls a web app and generates Playwright Page Objects and tests with an LLM.
- harness: runs local open-weight models as coding agents. Validation happens outside the model.
- gov-design-demo-simple: the Czech Gov Design System on Nuxt 4, without hydration mismatches.