Innings 3–5
May 27 → May 22 · 5 days earlier
6m · 2h service · 7m · 17m buffer
Scheduling-optimization SaaS for service businesses, approved on the Jobber marketplace. Built end to end: the scheduling engine, integration layer, API, and frontend.
A Monte Carlo engine that simulates every MLB game thousands of times, batter by batter, to price the market itself and surface edges against sportsbook lines. An exact Markov solver checks the sim’s math.
Private repo — code available on request.
A fan-made site from a lifelong Mets fan, tracking SNY broadcaster Steve Gelbs’ ballpark hot dog ratings. Community voting, video clips, and a share card drawn live from the standings.
Innings 6–7
Inning 6 · Building
Patcher's core is a multi-phase candidate pipeline: gap detection, pruning, Haversine travel-proxy pre-ranking, then travel-verified matching against each business's own rules. Five separate triggers used to race each other; they now feed one priority queue with request coalescing and distributed leases, so runs never overlap.
Under burst webhook load, a sync Supabase client wedged — one long-lived HTTP/2 connection stalling every request behind it. I traced the hang, migrated the hot paths to async with direct asyncpg, and collapsed 2–6 sequential round trips per request into single JOINs.
Inning 7 · Guarding
I built an autonomous workflow that pulls a Jira ticket, plans the change, and spins up subagents to implement it — a builder validated by a critic that drives the real app through Playwright. It maintains the e2e suite, opens a draft PR, and hands review to a separate model. Every stage runs together or in isolation.
As the solo SDET for a 10–15 person team, I owned the quality of everything we shipped. The result: ~1,000 Playwright end-to-end tests running in GitHub Actions in under 45 minutes through parallelization and Docker, plus ~1,000 backend and API tests in Python (Pytest and Behave) — all gating deployments from inside their project pipelines.
Inning 8
The release log — what shipped, season by season.
Scheduling engine, integration layer, and async infrastructure — built end to end.
Jobber marketplace · approved
Monte Carlo engine pricing every MLB game, 10,000 simulations at a time.
Community hot-dog rankings tracking SNY's Steve Gelbs, ballpark by ballpark.
Full-stack product features and the test walls behind them — ~2,000 automated tests currently gating production deploys.
