Engineering systems at Reward Gateway.

My responsibility changed over the course of my employment. I began by building the quality engineering function and its delivery systems. Later, I returned to direct technical work on production-incident analysis and release readiness for two AI products.

This page selects the work most relevant to the projects I take on now. It is not a complete account of the organisation or its product.

I joined the AI work after the products had been built. The immediate question was not whether a demonstration worked, but what evidence supported a release. For a multi-tenant retrieval assistant built with AWS Bedrock and OpenSearch across web, iOS and Android, I mapped the path from content publishing and ingestion through retrieval and streamed generation, then separated what could be tested from what remained blocked or unknown.

I built 180 tests across unit, integration, API and functional layers. Forty-four addressed security: authentication, guardrails, personal data, prompt injection and tenant isolation. Thirteen focused specifically on tenant isolation, because a cross-tenant disclosure would be a critical failure. I also built benchmark answers reviewed by subject-matter experts; a system whose responses vary cannot be evaluated usefully through exact-string equality alone.

The suite did not turn uncertainty into certainty. It made the remaining uncertainty visible. The content lifecycle was not covered end to end, the suite was not integrated into continuous delivery, response time had no agreed baseline or service level, and there was no labelled dataset against which to measure answer quality at scale.

I recorded each gap in the go/no-go report as a residual risk rather than allowing the test count to imply more confidence than it supported. The decision belonged to the panel. My responsibility was to make sure it was informed rather than surprised.

I designed the automation and measurement systems used by the function. They coordinated roughly one thousand end-to-end tests across pull-request, release and nightly runs, with the faster checks placed where they could inform a delivery decision.

I also built a library of more than seven thousand cases and linked structured test evidence to the changes being released.

I built a Python pipeline across the company’s incident tracker, documentation and monitoring systems to analyse 451 production incidents over ninety days.

The analysis found that 61 per cent could have been prevented through testing and that root cause was recorded on fewer than one in twenty incidents. It supported a £227,000 annual business case for preventable failure. This was a costed business case, not a claim that the amount had already been saved.

I built the quality engineering function from zero to fifteen engineers across London and Sofia, supporting more than two hundred people in product and engineering.

I also designed its career framework so senior engineers could progress without leaving technical work and ran an intern programme.