Pure mathematics BS, Rice STEM MBA (May 2026), based in Houston. I build energy market models that are tested the way a desk would use them: walk-forward, against the number the desk already has, with no look-ahead.
Energy:
- henry-hub-storage: weekly EIA gas storage nowcast from weather and power burn, scored against the analyst consensus over 231 weeks and tested on the futures reaction around the release; the model is frozen with a timestamped live record.
- energy-balance-signal: US crude and distillate weekly balances from EIA and CFTC data, aligned to release dates, and a walk-forward nowcast of Valero's reported refining margin from the Gulf Coast crack, with the interval reported even when it crosses zero.
- wind-performance-analysis: lost energy attribution on 420,480 SCADA records from a real wind farm; 464 MWh split by driver and turbine, a P50 budget with a variance bridge, and a live dashboard.
Other quantitative work:
- factor-ml-equity: rolling out-of-sample LightGBM on the US cross-section, 725 OOS weeks, with the no-lookahead test that makes the number mean something.
- nvda-supply-chain-nowcast: point-in-time test of Taiwan supplier revenue vs NVIDIA guidance; 79 SEC 8-K first prints, walk-forward over 46 quarters, look-ahead tests, pre-registered event study.
- carwash-credit-did: a pre-registered model race across four DiD estimators, selected on placebo behaviour, never on the treatment estimate.
- evo2-expression-pipeline: sharded, resumable inference on a 7B genomic model and a systematic regressor comparison.
- ai-research-talent-map: 109,378 faculty publication records mapped onto seven AI research areas, with every total reconciled to source rows.
Open to energy trading analyst, quant, and origination roles in Houston or New York.