Multi-vendor checkout
One cart, several vendors, everyone paid correctly.
Open to roles · Rochester, NY or remote
Adrian
Marcellus.
Full stack engineer.
Full stack, from schema to QA. Every workflow tested in a real browser. AI agents that follow my process, not the other way around.
Selected work
Things I built and shipped.
Two projects, shown through recordings of them actually running. The details stay private; the engineering doesn't.
Testing, end to end
If it isn't tested, it isn't done.
Testing is where I spend the most care. The suite goes from pure functions up to full browser workflows, and it records every run, so the demo and the proof are the same footage.
How the suite is layered
Fast at the bottom, real at the top.
Most checks are quick and precise. The ones at the top are slow on purpose: a real browser doing the whole journey, recorded start to finish.
- End to endA real browser doing what real people do, on data it made itself.
- API & guardsEvery door checked for who's allowed through.
- ComponentUI pieces mounted on seeded data and driven on their own.
- UnitPure logic and the math that has to be exact.
01
Every run is recorded
A failure comes with the full story: a recording of the run, which data checks broke, and what was expected next to what actually came back.
02
Done means proven
Each feature says up front which checks prove it works. An audit compares that list to the tests and flags anything promised but not proven.
03
Regressions get caught
Every fix gets a guard. Whether a person or an agent wrote the change, anything that brings an old bug back fails the build before it ships.
04
Stages build on each other
Tests run in the order a real user would go, each one starting from what the last left behind, so the whole journey gets exercised.
From the recordings
Every clip here is a test run.
Nothing staged and nothing seeded. Every record on screen was made earlier in the run by the suite clicking through the app, then used again further down the line. Only payments run in test mode.
Discount codes
Every kind of code prices the order the way it should.
Vendor approval
A rejected profile gets fixed, resubmitted and approved, then connects payouts.
Order cancellation
Cancelling puts everything back where it belongs.
Fulfillment
Shipping an order stays consistent from label to doorstep.
Event lifecycle
Create, publish, edit and expire, everywhere it appears.
Strong fundamentals, newest tools
Classically trained. Now I run agents like a team.
Page Parrot was written by hand, every line. That's the background I bring to today's tools. I use agent harnesses for more than autocomplete: the repo carries a playbook of skills, rules and commands I wrote, so agents follow my process, and I can read and check everything they produce.
From idea to recorded test
Every feature goes through the same steps, with a human sign-off between each.
Bug loop
When a test fails, one agent decides whose fault it is. A wrong test goes back to be fixed. A real bug gets written up and handed to a separate agent, which isn't allowed to bend the app to satisfy a bad test.
failtriagereportfix
Code-reading audits
Agents read the riskiest parts of the code piece by piece against the rules they're supposed to follow, and file what they find for review.
splitreadfilefix
Rules as guardrails
Project rules stop agents from repeating known mistakes, like fixing a test to match what was intended rather than just to pass, or asking instead of guessing.
ruleagentcheckship
- Agent harnesses
- Agent skills
- Parallel subagents
- MCP
- Playwright
- Sentry
Experience
Where I've worked.
2025 – present
Full Stack Web Developer
JPTabb & Co. · On-site
- Main developer of a multi-tenant business directory and marketplace: payments, dynamic shipping built on the three major carrier APIs, refunds, caching.
- Built the Playwright and Vitest suites that gate every release.
2024
Full Stack Web Developer Intern / Co-op
JPTabb & Co. · Co-op
- Main developer of Page Parrot, a live-chat widget integrated with the Slack and Discord platforms, with a Node REST API and Stripe billing. Written by hand.
2023
Software Engineering Intern / Co-op
Xerox · Webster, NY
- C# telemetry for print and scan workflows, reported through Azure Workbooks and KQL.
2021 – 2022
IT Technician
Skyport-IT · On-site
- Hardware, networking and end-user support.
Skills
Languages
- TypeScript
- JavaScript
- SQL
- C#
- Python
Frontend
- React 19
- Next.js
- Tailwind
- Radix
- TanStack
- Accessibility
Backend & data
- Node.js
- Postgres
- Supabase
- Redis
- REST
- ETL
Testing
- Playwright
- Vitest
- E2E design
- CI gates
Infra & payments
- Vercel
- Docker
- AWS
- Azure
- Stripe Connect
AI
- OpenAI
- Anthropic
- Entity resolution
- Vision models
Education
B.S. Computer Science
Rochester Institute of Technology · 2024
A.S. Computer Science
Finger Lakes Community College · 2020