I build the verification layer for AI-assisted software.

Self-triaging incident pipeline

Production errors classify themselves, draft their own fixes, and open confidence-gated pull requests. The confidence gate decides what's safe to open for review. Humans approve every merge; the pipeline just does the legwork.

Impact: triage now lands as a ticket and a draft PR within minutes, before anyone opens Sentry

Coverage that grows itself

Every closed production bug becomes a Playwright regression test via Claude Code, driven by a tag-based state machine on a weekly schedule. The suite widens on its own, so yesterday's incident can't reappear unnoticed.

Impact: a new regression test generated for every closed production bug

Parallel E2E on shared state

Parallelized Playwright across a single shared-state environment using tag-based serial lanes (@readonly / @stateful) and per-worker storage state. That cleared the race conditions on a non-isolated backend, and it did it without a fresh fixture for every test.

Impact: 50% reduction in run times on parallelized tests

Quality, built from zero

Stood up the frontend unit-testing practice from scratch with Jest + RTL + MSW, enforced in CI so a failing test blocks the merge instead of slipping into a release. Wrote the testing standards the team builds against today, as the first QA voice on the product.

Impact: 40+ component tests in CI, and a failing test now blocks the merge

QA that writes itself, on demand

Authored custom Claude skills and slash commands that generate bug-fix test coverage across frontend and backend. The repetitive QA chores are now a one-command workflow anyone on the team can run.

Impact: new coverage that took a day of QA time now ships from a single command

FDA-grade V&V

Owned formal validation, verification, and traceability for FDA-submitted SaaS algorithms: STD / STR / RTM documentation backed by automated GitLab CI/CD tests. Designed the Java and Selenium framework that a distributed team of eight built on.

Impact: supported FDA submissions with full requirement-to-test traceability

Pipeline: Hebrew video editing in one click

Upload a raw Hebrew talking video, get back a polished cut: word-level captions (Whisper + a Claude proofread pass), silence trimming, audio cleanup, AI 4K upscale, stock B-roll, and one-click scheduling to socials. Python GPU backend on Modal, Playwright tests on the UI. It's invite-gated: create a user, then ask me for an invite code.

Egg Smash Adventures

A casual mobile game I built solo and shipped to Google Play. Claude + VS Code for the build, Supabase for auth, accounts, and cloud save. And yes, GitHub Actions runs daily sanity tests on it, because apparently I can't ship anything without a pipeline.