Medicine, science, software.

I trained in medicine at the University of Cambridge, where I also completed my PhD. Those years taught me to think clearly, question deeply and care responsibly. They also left me interested in what happens between a good clinical decision and the systems needed to carry it through.

I taught myself to code in 2006 and helped start a Linux society at school by 2009. I have contributed to open-source software ever since. Coding was never separate from the rest of my interests; it became the way I investigated difficult technical problems and made ideas concrete. That habit has stayed with me: understand a problem end to end, write the code and remain responsible for making it reliable.

My work now sits in primary care, close to the everyday complexity of real lives. I work in a small engineering team as a founding-style engineer with end-to-end ownership of clinical AI products used at national scale. I own Requests, and spend most of my time close to the code: defining problems, choosing the architecture, shipping systems and improving them once they meet real clinical workflows. This includes the patient questionnaire engine, Smart Triage, clinical document automation and our EMIS Web and SystmOne integrations.

The work is direct and hands-on rather than peripheral: writing the code, seeing how it behaves in practice and owning the fixes and iteration that follow. The aim is clinical AI that turns patient requests, documents and the signals inside them into useful action. I care about technology that respects clinical judgement, fits the way teams actually work and earns its place through usefulness, transparency and trust.

I have always liked ideas that must survive contact with reality. While at Cambridge I founded and built MayBall.com, taking its ticketing, marketplace and event operations from a blank page to a platform that processed more than £4 million. It was my first real education in product, reliability, commercial responsibility and the unglamorous details that make a system work.

That engineering thread also runs through open-source machine-learning systems used far beyond my own projects, including merged work in TensorFlow Models and Tensor2Tensor. Those contributions were specific rather than grand, but they reinforced the value of clean interfaces, careful abstractions and code that makes the next person’s work a little easier.

That engineering work sits alongside my research. I have used neural networks to analyse biomedical data in Nature Methods, contributed to a randomised clinical trial published in Nature Communications and to work in Blood on directing stem cells towards platelet production, among other papers. I have also trained and presented a billion-compound AI model at NVIDIA GTC.

I write to think in public. The essays here range across medicine, machine learning, research and the behaviour of complex systems. They are working notes rather than a finished doctrine: a record of questions I think are worth taking seriously.