About

Most of what I know came from taking things the whole way. I own the engineering for TubeMate, a live AI product with real users and a billing system, which means I have had to care about the parts that are invisible when a product is going well — rate limits, abuse controls, error tracking, and finding out that a metric I had trusted for months was measuring the wrong thing.

Through SOLAT I deliver client systems: a structured content platform with a nine-role permission model, a booking product whose availability engine has to survive British Summer Time, and a retail storefront where the interesting decision was declining to build a checkout. Those are written up in client work.

I am also doing early research into speech recognition for Nigerian languages through ÒreAyò. It is genuinely at an early stage, and I would rather say that than dress it up.

How I work

I try to pick the architecture a problem actually needs rather than the one its category expects. The decision I am asked about most is choosing long-context prompting over retrieval for TubeMate, because retrieval solves a problem I did not have and would have added an index, latency and a new way to be confidently wrong.

I would rather ship something small that works and can be explained than something broad that cannot. That extends to how I write about my work: where a number is not reliably measured I say so, and where a system falls short I would rather be the one to point at it.

Alongside this

I currently work in a regulated custodial environment in the UK while building my technology career. It is not technical work, but it has sharpened how I approach responsibility, documentation and decision-making under pressure.

Where I’m heading

I’m looking to develop deeper engineering experience inside a strong technology team, while continuing to build and learn through TubeMate, ÒreAyò and selected SOLAT work.