Ideas
Research notes, projects and experiments. Newest first. The map shows how they connect.
Machine learning is function approximation
A model is a function you cannot write down, so you build one out of simple parts and tune it until it fits. An interactive walk through the standard parts, and what each is good and bad at.
Leaving JetBrains
My internship on MCP Steroid has ended. What I learned at JetBrains, who I learned it from, and what comes next.
Joining JetBrains
Starting in the Agentic Experience & Tools team, on MCP Steroid and local AI.
Taking Soqra to a classroom
A school visit with 10th graders, and what 93% of them said about learning maths with AI.