Hiba R.

I build test automation that catches real bugs, not suites that go green while production breaks. Nine years as an SDET across fintech, insurance and AI platforms, most recently testing AI agents and LLM pipelines. What I do: - Build Playwright, Cypress and Selenium frameworks from scratch, in TypeScript, Java, Python or C#. Not scripts dropped into someone else's scaffolding, a structure your team can actually maintain after I hand it over. - API and contract testing with Postman, REST Assured and schema validation, including the service-to-service boundaries where most integration bugs hide. - CI/CD quality gates in GitHub Actions, GitLab CI and Jenkins, so broken code stops at the pipeline instead of at your users. - AI and LLM testing: LLM-as-a-Judge evaluation, golden dataset scoring, hallucination and quality-drift tracking, and agent trajectory failure analysis. - Performance testing with JMeter, and accessibility testing to WCAG as part of release validation, not as an afterthought. Where I have done it: - Currently SDET Lead AI at Afiniti, where I designed the QA strategy from scratch for an AI orchestration platform built on FastAPI and LangGraph, and built a 12-agent autonomous QA framework with human approval gates. - Senior SDET AI at Beam AI in Berlin, owning quality for a no-code agent automation platform, including event-driven workflows and billing pipeline data integrity. - Senior QA Engineer at Finlex in Frankfurt, a regulated insurance SaaS platform, where I was the release gate and built audit-ready traceability. Also built payment test frameworks covering ISO 8583 and ISO 20022 message validation. Something I built on my own time: OAQA, an agentic QA framework that generates Playwright tests, repairs them when selectors drift, and triages its own failures to separate genuine regressions from flakes. How I work: I ask what breaking actually costs you before I write a single test, because a checkout flow and a settings page do not deserve the same coverage. You get evidence-backed bug reports with payloads, screenshots and recordings, not "it doesn't work". Happy to start with a small paid trial task so you can see the work before committing to anything larger.