Lines of work
numeraiTournament forecasting pipelines — feature engineering, ensembling, and signal that survives out of sample. Python · Rust
calibrationTooling to score predictions over time: Brier, log-loss, reliability curves. A forecast without a track record is a guess. research
belief graphsCertainty as a first-class value — strength/confidence pairs propagated through a reasoning graph. Elixir
The stance
Most “AI prediction” skips the only part that matters: keeping score. Pythia treats every forecast as a claim with a deadline and a ledger. The interesting work isn't the model — it's the apparatus around it that tells you, ruthlessly, whether to trust it.
Forecasting, calibration, or quant ML?
Open to collaboration and contract work in this space.