Marius Argatu

- I built an internal debugging agent powered by LLM that connects the failing test output with Datadog traces, the current stack trace and the recent Jira changes, and then it loops until the test passes or a real bug appears. This reduced the bug triage time from hours to minutes.

- I own the quality strategy inside my feature squad, which means I decide what to test, on which level and why. The developers write their own unit, integration and system tests, and I review these tests and advise them about coverage gaps and about where each test belongs in the pyramid.

- I maintain and extend the API test framework with new fixtures, helpers and reusable modules, and I build coverage for the new payment endpoints and business flows.

- I apply a shift left approach in practice. During the low level design phase I write the test scenarios, and these scenarios become the specification for the tests that the developers write.

- I use agentic workflows every day, for example MCP servers for knowledge base access, slash commands for bug triage, and agents for test maintenance.

- I am responsible for the regression planning and the release gate decisions in the payments domain.

- I built the LLM evaluation for a RAG assistant that answers researcher questions based on FDA and EMA guidance and on tenant submissions. I separated the retrieval quality (hit rate, precision@k) from the generation quality (faithfulness, answer relevancy, using RAGAS/DeepEval). The golden datasets were stored in Langfuse and the evaluation runs were executed in CI.

- I built a model based testing system for the permissions matrix using xState. From a single state machine definition it generated around 1,000 test scenarios, and it follows the product when the permissions change, without any manual rewrite.

- I built a test platform in Playwright and TypeScript with POM and with isolation of tenant and role based on fixtures. It was adopted by several functional teams.

- I created a migration tool based on AST that translated a legacy Karate DSL suite into equivalent TypeScript and Playwright code, so a manual rewrite of thousands of lines was not necessary.

- I was the framework SME, responsible for onboarding, reviews of test patterns and coordination of cross cutting changes, including a rewrite of the authentication system.

- I built a custom Pytest plugin that runs dependent API tests in parallel, with the order resolved by a DAG. This reduced the runtime and kept the correctness.

- I patched and rebuilt the Chromium fork of Qt WebEngine to expose the remote debugging port, so I could replace Appium with Selenium through CDP. The effort per test dropped from 4 days to 1 day.

- Automated Integration Tests