I build test suites that catch regressions before release — and AI tooling that helps find them faster.

Playwright + TypeScript Test Framework

250+ UI and API tests with custom fixtures, typed page objects, a @Step decorator for Allure reporting, and CI that publishes a live report to GitHub Pages on every push.

An AI layer built on top: an MCP server and four agents — a requirements analyst, a failure analyst, an adversarial judge that reviews the failure analyst's output, and a spec generator that produces tests from requirements and refines them in a heal loop until they pass.

LLM Test Harness

A five-week test harness for non-deterministic LLM outputs. Property-based assertions, pass-rate thresholds, adversarial inputs (prompt injection, role confusion, payload splitting), rubric-based LLM-as-a-judge, a real RAG retriever (Qdrant + Ollama embeddings) with faithfulness, relevance, context precision and recall metrics, and streaming tests (TTFT, mid-stream disconnect). Documented in a week-by-week log with a published Allure report.