Work

Lightworks, Scotiabank, Sun Life, and TELUS build and govern the Agentic Control Plane together, and every member holds perpetual rights to what we create. I lead TELUS's engineering contribution.

The platform is the control layer for agents, models, users, and inference pipelines: approval workflows, policy enforcement, machine identity, and audit trails that let a bank or an insurer put agents into production without losing sight of them. It already runs at production scale across the Consortium.

The Sovereign AI Factory in Rimouski is TELUS's flagship sovereign compute facility, fully operational since September 2025 and sold out.

Rimouski is the proof point for the national expansion announced with the Government of Canada, Province of British Columbia, NVIDIA, BC Hydro, and Westbank. Kamloops, Vancouver M3, and 150 West Georgia come next. The purpose is consistent across the network: Canadian AI workloads processed, stored, and managed inside Canada.

wâsikan kisewâtisiwin — "kind electricity" in Cree, named in ceremony — is Shani Gwin's company and creation: an AI trained in-house with Indigenous elders and community members that gives authenticated, Indigenous-informed guidance to anyone writing about or researching Indigenous Peoples in Canada, correcting in real time and auditing documents for bias.

The pilot kicked off in January 2026 and was announced that May, making TELUS the startup's first beta tester. I led the work: the model runs as its own deployment within Fuel iX — selectable when teams build copilots — and is hosted on the Sovereign AI Factory, keeping it on sovereign Canadian infrastructure in an OCAP®-friendly environment.

More than 300 team members joined the formal pilot, thousands more have access, and I proposed the joint purple-team effort helping the wâsikan team harden the model ahead of its public release.

Unicorn.ai had proved demand inside TELUS. The next problem was fragmentation: every new AI product needed models, company context, tools, identity, moderation, observability, and cost controls while the underlying technology changed by the month. Nobody wanted to rebuild that stack once per application.

I led the technical work of separating those concerns into Fuel iX, and the contributors who had leaned into Unicorn became its first full-time team. Applications could change models or clouds behind a common gateway; access, policy, orchestration, and telemetry stayed in one runtime.

Unicorn.ai began on March 26, 2023 as 65 lines I wrote over a weekend. By the end of the day it was running in Slack and on the web. What made it interesting wasn't the chat box. It was where the AI sat: inside conversations where work was already happening, reading the same thread as everyone else, with its knowledge, costs, and actions in the open.

I developed it in the open and pushed code daily, partly to build the thing and partly to show how fast we could ship inside a company of TELUS's size.

Katie Peters, a staff developer on my team, championed a clear vision: focus, flow, and joy for developers. As head of engineering productivity, I led the enterprise adoption that made GitHub the single source of truth.

We rolled out push protection enterprise-wide in one day. It was non-blocking, allowing developers to acknowledge a finding and continue. We received zero complaints. Secret scanning remediated thousands of accidentally shared secrets, while GitHub Actions automated testing and validation.

Quality was a downstream gate when I arrived: a small testing group, long feedback cycles, and a growing browser suite expected to catch problems after development. Automating that queue would only have made it bigger. We moved quality into development, then followed the constraints that surfaced in tooling, delivery, infrastructure, and production.

Six days after I started Bueller, the same grocery suite ran successfully against a second Loblaw banner by changing only its root URL. That small result proved the design: the customer journey could remain the same while the brand, language, environment, browser, and device changed underneath it.

Bueller became Ferris and expanded beyond grocery into fashion, beauty, loyalty, and pharmacy. The common machinery was extracted into a shared library while individual products developed their own suites and maintainers. The practice moved below the browser as well, adding reusable API clients and self-service tools for creating users, carts, orders, and other product state.

The bigger change was ownership. Other engineers took responsibility for individual products, extended the shared system, and replaced parts as the technology changed. I spent less time writing automation and more time building the teams around it.

When I founded Platform Engineering, Loblaw Digital's products were split between data-centre infrastructure and a growing collection of cloud experiments. We built a common operating platform so product teams did not each have to invent how to provision, deploy, secure, and run their services.

The first major proof was online grocery. Working with Google Cloud Professional Services and Publicis Sapient, we moved its existing SAP Hybris system to Google Cloud in six months. The initial lift-and-shift was deliberately pragmatic, while its Terraform foundations and architecture set up the later move to Kubernetes.

That production migration made cloud part of the normal path from a code change to an owned service. Platform Engineering connected infrastructure, delivery, observability, and reliability around the teams responsible for the products.

Loblaw Digital had more than 150 web experiences to modernize, but creating one meant navigating Windows servers, central infrastructure, and hand-operated releases. I co-founded the Digital Experience Studio to replace that delivery path, not simply push the same backlog through it faster.

Small, autonomous teams brought product, design, content, and engineering together around reusable components, starter kits, and a shared delivery platform. Preview environments and static delivery on Netlify removed the application-server queue, so a team could take an experience from idea to production on its own.

Iris grew out of Jeanie, a Ruby bot I built after noticing that PC Express colleagues were already requesting operational changes in Slack, then waiting for someone with access to the right system to carry them out. Jeanie proved the idea; Iris was the Internal Tools team's ground-up rebuild in Elixir.

The team turned a collection of commands into a permission-aware platform with interactive forms, multi-step approvals, persistent sessions, observability, and a modular architecture other development teams could extend.

Through Iris, field colleagues could correct product listings and adjust store operations from their phones. Employees could request access, while engineering teams could prepare releases and coordinate incidents without manual handoffs.