The work

An internal notifications platform, roughly six months old when I joined and built out across the four years that followed, handling tens of millions of merchant communications. Service-oriented, event-driven, with the delivery guarantees that implies: retries, ordering, idempotency, and the operational resilience to survive the year’s peak traffic.

Full-stack across it - the front end, the services behind it, the data stores underneath, and the observability over all of it.

What I was responsible for

• Technical leadership: architecture, scaling and incident response
• Driving the migration that moved the rest of the business onto the platform rather than going direct to the notification senders, including shaping the API and the developer experience around it
• Scaling for Black Friday across several years, including a sixfold increase in one of them, with no downtime in any
• Building incident practice into the team, focused on early communication and getting the right people involved quickly
• Root cause analysis across the stack:
  • A thundering herd on pod startup exhausting TCP accept queues, fixed with the ActiveRecord schema cache
  • An HMAC hash sitting in a hot code path, accounting for roughly half the workload
  • Campaign loading, from timing out to sub-second
• Replacing VCR cassettes with service emulators, so tests ran against the protocol rather than against a recording
• CI checks that made whole classes of error fail the build
• Zero-downtime data migrations of hundreds of millions of records, staged so the business could hit decommissioning targets
• Mentoring, and running knowledge-sharing sessions

AI-assisted engineering

• Defined the team’s MCP tool standard - how tools were structured and implemented - and built the first tools on it so the rest of the team could keep going
• Built an AI-driven debugging workflow using MCP servers wired into observability and analytics, which used critics, hypotheses and facts to rule out dead ends early, and shared it across the wider company
• Raised duplicated effort at company level: across thousands of engineers the same problems get investigated, written up and reviewed repeatedly, and nothing retains what was already learned