work
Current projects, experiments, and the public work behind them.
Current focus
Tuned Tensor
A feedback loop for open-weight models: define the target behaviour, fine-tune, check for regressions, and turn failures into the next dataset.
Website · Email-triage case study · Dataset · Model · CLI
Selected experiments
Deeper Level
Play chess against a model, then inspect the candidate moves, threats, plan, and critique behind its reply.
Autoresearch Classifier
An experiment loop that searches for better prompt-injection classifiers using scikit-learn.
Mental Model Graph
Explore how mental models from economics, psychology, mathematics, and logic connect.
Still Here
A reflective web app for counting days and writing about the ones that matter.