Ash Street Senior frontend developer in Toronto

About 250 engineers across 500+ repositories shipped through the GitLab CI/CD platform I designed.

Builds were taking around ten minutes, which is a long time to sit and watch when 250 people are doing it all day. The problem turned out to be memory: the runners just weren’t big enough for the jobs. Giving them more made an immediate difference, and refactoring our Dockerfiles to cache properly took it the rest of the way. Under three minutes.

The point of that wasn’t really the time saved. Teams could deploy whenever they wanted instead of batching work up for a big release, and when a build did fail it failed early, so it got fixed faster too.

I also set up the React/Next.js and TypeScript templates new apps start from, so a new project arrived with the structure, testing and build config already in place.

GitLab CI/CD, Docker, Kubernetes/EKS, AWS, React, Next.js, TypeScript

Moving an entire enterprise GitLab instance to another one, 457 repositories, without breaking the pipelines, permissions and branch rules attached to them. Two of us did it.

The migration, and the management of the whole instance, ran on scripts I wrote. Repodex is the GUI for running those scripts: transfers, audits, drift checks and reports in one place instead of a folder of commands. It also worked out the permissions each repo needed in the right shape, which is the part that does not survive being done by hand.

Python, Flask, GitLab · Source on GitHub (opens in a new tab)

MyTELUS is where TELUS customers manage their own accounts, and billing is one of the main reasons they sign in. Internally the figure was about $17 saved every time someone sorted something out in the app instead of phoning in, which is the whole argument for self-serve.

Billing wasn’t easy to find, and it wasn’t as accessible as the rest of the site. I refactored our BFF so it returned the billing data in one reusable shape, then built a new, accessible billing card on top of it. One component covers every state a customer can be in: paid, due, overdue, on autopay, or looking at a first bill with a part-month on it.

React, TypeScript, Node.js BFF, GraphQL

{status: "due",balance: 84.2,dueDate: "2026-03-14",daysLeft: 6,autopay: false,firstBill: false,}

A business initiative was being scoped at more than $2 million, and the case for it rested on analytics coming out of the app I worked on.

Our product manager couldn’t work out why one value was so high and wanted a developer to dig into it. I traced it back through the app and the data flow and found the value was hard-coded in brackets, data['value'], instead of the parameter being passed in, data[value]. Every call returned the same field, whichever metric you asked for.

I fixed it and stayed on it with the analytics team until they confirmed the corrected numbers in production. With the real numbers in front of them, the business cancelled the initiative before it kicked off.

Analytics, production debugging

A rebuild of the public site for a loyalty programme most Canadians have a card for. Six of us, six months, with a compliance deadline for accessibility on the end of it.

I was the hands-on tech lead: the React component strategy and library, the Node.js backend-for-frontend, and the OAuth login end to end. Accessibility went into the components themselves, so nobody had to audit it back in at the end. That’s how we hit WCAG AAA before the deadline.

I pushed back a lot on colour contrast. Getting design to accept that half the brand palette stops working once you’re at AAA was the longest argument of the project.

React, component library, Node.js BFF, OAuth, WCAG AAA

Nissan’s car configurators were moving off a legacy CQ5.5 setup onto newer AEM with React, while staying live for people picking a car.

I owned the frontend of the Kicks configurator, including the vehicle-data APIs behind the options people click through. It shipped to WCAG AA and was tested with Jest.

React, styled-components, AEM, AWS Lambda

I play centre in Toronto United Flag Football. The league kept its stats in a 20-column table that nobody could read on a phone.

I built player cards you can search and sort instead, backed by a small Node API that normalises each league’s stat names into one shape. It’s tracking 172 players this season. It’s grown into Afterwhistle, so adding another league is a config change instead of a whole new fork.

Open tuff.afterwhistle.ca (opens in a new tab)

React, TypeScript, Node · tuff.afterwhistle.ca (opens in a new tab)