I suddenly became very sick in the second year of my undergraduate degree at Brown University. Ten years on and I have a grab-bag of partial diagnoses, a smattering of weird test results, and no real answers. I was so disappointed to learn that being mysteriously sick in real life isn’t like TV, that it’s not like a cartoon where the villlain is thrillingly revealed at the end and subsequently vanquished. Often doctors wanted to help but were at a loss, and I was left to try to manage things best I could on my own.

I’ve been fortunate to encounter a few really incredible doctors on my medical journey who made out-of-the-box diagnostic or treatment suggestions that changed my life. Finding those doctors was difficult, time consuming and expensive. It pained me to think of how few people would have the time and resources to keep looking until they got answers. It pains me still to think of how many people with malfunctioning bodies like mine are living extremely limited lives when there is information out there that could help them, if only they could access it.

The doctors who helped me didn’t have some magic insight – they didn’t understand quite what was wrong with me either. What was different about those doctors was that they had seen other patients who exhibited my particular symptoms before. We tried out the interventions that had worked for those patients, and they happened to work for me too. We don’t know why they worked for those patients, and we don’t know why they work for me, but we can see that there is a pattern. In other words, the novel thing that those doctors had was not knowledge – it was data.

I do machine learning, so data is my second language. I started to think about the unique data those doctors had. I started to think about what sorts of algorithms could mine that data, and how we could democratise access to it so that you didn’t have to see twelve private specialists to learn something useful. I started to daydream about a future in which patients with the same symptoms could find each other and compare notes, even if no one could diagnose them. I started to wonder if we could generate research datasets so that one day someone could study us, and not just find a pattern but really understand what was wrong and figure out how to fix it, so that in the future the doctors who wanted to help would have the information to do so. I realised that the tools to do this already existed and were right in front of me. And so Celie was born.