Applied analytics · intelligent systems · probabilistic decision automation
I design and operate production decision systems — the kind that ingest noisy real-world data, turn it into well-calibrated probabilities, and act on them under risk controls. End-to-end ownership from data ingestion through automated action and post-hoc P&L attribution.
Live algorithmic trading system on Kalshi's CFTC-regulated daily-high-temperature prediction markets across 20 US cities. Real money, maker-only execution, behind a 15-minute Telegram approval gate.
- Six-model probabilistic ensemble — NBM, GFS, ECMWF 51-member, HRRR, NWS gridded, MOS
- Per-(model × station) EMOS calibration — Platt scaling fit on a 152K-row two-year backtest; +26.7% Brier reduction on holdout
- Quarter-Kelly sizing — disagreement-weighted multiplier, soft-drawdown derisking, asymmetric edge gates, 50¢ side-cost floor derived from P&L attribution
- Maker-only execution — limits at
ceil(ask × 100) − 1for 0% fees; single-instancefcntllock; fill-watch re-quoting - Backtest-gated promotion — every production change validated against historical data before it goes live
→ Source: GEO-The-Kalshi-Weather-Bot-DEMO
→ Live demo: apeabody007.github.io/GEO-The-Kalshi-Weather-Bot-DEMO
Python 3.11 · SQLite · NumPy / SciPy · REST APIs (incl. RSA-PSS authenticated trading endpoints) · launchd · pytest + AST-based invariant tests · Playwright
Data feeds: NOAA GHCND · NWS api.weather.gov · Open-Meteo (ECMWF / GFS / HRRR / NBM) · METAR · NWS Climate Reports · NWS MOS
- Backtest before promote. Every change to filters, sizing, or model weights gets validated on historical data before it touches live capital.
- Brier improvement ≠ trustworthy calibration. Always audit the shape of a fitted curve, not just its score.
- The crowd is usually right at extremes. 1¢ and 99¢ markets carry real information; encode the skepticism as a hard veto, not a soft filter.
- Encode expensive lessons into filters or tests. Memory is lossy; AST-pinned invariants and dated investigation memos aren't.
BA Economics — University of Wisconsin–Madison BS Psychology — University of Central Florida
I think in systems, behavior, and incentives. Equally comfortable with the mechanics of an ensemble forecaster and the second-order effects of how a system's design changes the behavior of the people who interact with it.