Scope
one problem, named
founder decides, agents build
AI Platform Engineer · Fintech Operations
At 2am a SWIFT MT103 batch failed with 4,112 records held. For years a human paged on-call, found the poison record, resubmitted the batch, and released settlement by morning with no loss and no sleep. The same job now runs as agents that isolate the record, resubmit, and settle while the human stays asleep.
01 / About
I am an AI platform engineer at a global bank in Montreal, supporting enterprise deployments of Claude, ChatGPT, Grok and Copilot. I came to it after ten years keeping bank systems running, the last five in payments.
I studied electrical engineering at McGill and planned on control systems, but found myself at HSBC on market risk systems instead. I spent nearly five years in Toronto supporting the platforms that run value-at-risk and Monte Carlo simulations, in a follow-the-sun team across four countries. At Société Générale in Montreal I moved to payments. I supported the SWIFT, Montran, ACH and Flexcube applications, which handle thousands of transactions a day worth millions of dollars, in a regulated correspondent bank. I led testing and deployment of the ISO 20022 MX migration across more than five business units, and it went live with zero rollbacks.
Across the two banks I shipped 134 permanent root-cause fixes and mentored five analysts. Outside the bank I run The ARCHV, a football history brand built on a production system I designed myself: over 120 Claude skills, scheduled agents for research and editing, and a native iOS app. Same discipline, different system.
I grew up in Lagos, moved to New Hampshire at nine, and came to Canada for McGill. My partner and I run our household on an app I wrote for us, and outside work I make music in Ableton Live and engineer the audio for a friend's podcast, Pros in the Arena.
I write down what I learn in the field notes, for anyone moving into this work and for anyone who wants to be happier in the job they already have. For one sourced AI story most days, follow along on Threads.
02 / The delivery
Scope to Run, the way I actually ship. Scroll through it, or drag the timeline.
one problem, named
founder decides, agents build
The ARCHV iOS app, build 40
agent fleets, every diff reviewed
238 tests, none failed
fresh-install UI gates
review caught shipping bugs in 4 of 6 builds
deploys verify themselves before going green
a desk that publishes every morning
health probes that email before users notice
Two live products, both built and run end to end. One tells football's story with an iOS app live on the App Store, and one explains AI news in plain English.

An immersive archive of football's greatest moments, retold across social and in an iOS app that is live on the App Store. The owned site, thearchv.ca, covers football, NFL, F1, tennis, golf and basketball in illustrated form, with long reads on the systems that run the game. The whole content operation behind it runs as a fleet of scheduled AI agents: generation, scheduling, analytics, and a performance log that breeds towards what works.

A Threads account that posts one recent AI story most days, in plain English and with its sources named. It runs on a scheduled agent that pulls the latest real stories, writes each post to a fixed brief, and checks every one against hard rules on sourcing and voice before it ships. No invented numbers, no hype.
This is how I size a two-week build. Answer three questions and the sketch fills in, honest about what fits in a fortnight.
Lived-experience write-ups on payment infrastructure and running AI agents against real work. Expanded from what I post each week.
A running digest of payments, agentic commerce, and the AI stories that touch how money will move. Sourced, edited, and written plainly.
Definitions and explainers for the terms the brief keeps using, each with its sources.
@thearchv.ai posts one recent AI story most days, in plain English and with its sources named.