Fellow · frontier AI evaluation
AI Trainer at Handshake AI
As a Fellow on Handshake AI’s program, I design hard coding tasks that frontier AI agents are trained and measured against, each with a container, hidden tests and a grader that holds up under review.
About
Handshake AI pays domain experts to help train and evaluate frontier AI models. As a Fellow on its Project Insight, I create coding tasks that need real domain expertise, built to find where today’s best AI agents still fail.
Each task is a complete package: a containerised environment, a precise brief, hidden tests that decide whether a solution is right, and a grader. The hard part is making a task difficult for the right reason, so it measures real skill rather than a trick, and making the grader hold up when it is reviewed.