Wingston Sharon Wilson
Can we measure what these systems actually do?
Nine years leading engineering teams at Booking.com, growing engineers and getting products in front of customers. The rest of my time goes on testing AI systems and training small models on my own hardware.
Amsterdam · open to conversations →
Testing AI systems
Running AI on your own hardware
Machine learning in production
Creative technology
Selected work
What I’m building
Agentosaurus
A site that helps people find groups working on climate. AI agents write most of its code. In 2026 they opened and merged over a thousand pull requests. They never message each other. They read each other’s code reviews instead.
Django · Claude Code · k3s · Own GPUs
Proving what a model learned from
If someone fine-tunes a model on your data, can you prove it afterwards? I plant keyed markers in the training set. Later you can test any model and prove whether it learned from your data.
Training data · Fine-tuning · Evaluation
Psychohistory
Three years asking whether you can forecast crowd behaviour like weather. Ongoing. Everything is public, including what hasn't worked.
Research · Open · Testable
Nine years leading engineering teams
Software engineer, team lead, senior engineer, then engineering manager across several teams. The job is mostly people: hiring well, growing engineers into scope they haven’t held before, and giving a team enough room to find out whether an idea works. In Accommodations I went from running one team to three, which meant learning to lead through other leads rather than doing the work myself.
Full story →Any of this useful to you?
I’m happy to talk about running AI on your own hardware, testing AI systems, installations, and workshops. I’m also looking for a new role.
Get in touch →