I build forecasting systems, and the validation tooling that decides whether they are any good. My main project takes broadcast video in and puts calibrated probabilities out across four sports. The part worth looking at is the harness: leak-guarded walk-forward validation, pre-registered claims, and an acceptance gate with the authority to reject my own results. It usually does.

Open to ML, Computer Vision, Quantitative Research, and Sports Analytics roles. US citizen, no sponsorship needed, available anywhere in the US or fully remote.

court-vision -- an end-to-end multi-sport forecasting platform (NBA, MLB, soccer, tennis).

Broadcast Video -> YOLOv8n detection -> SIFT homography -> Kalman+Hungarian tracking
  -> OSNet re-ID -> OCR -> court-coordinate features that exist in no public dataset
  -> calibrated win-prob + prop models -> possession-level Monte Carlo
  -> walk-forward, leak-guarded validation -> receipt-backed evidence layer

What makes it different is the epistemics, not the model zoo:

• Leak-free walk-forward validation with assertion-level per-fold guards and truncation-invariance property tests -- the harness fails the build on lookahead.
• 21,000+ pre-registered hypothesis cards: claim, expected sign, and magnitude written down before outcomes are looked at, then graded mechanically. The large majority of graded cards are honest REJECTs -- which is what real signal discovery looks like.
• A full-season market-efficiency result I publish instead of hiding: my calibrated models match the Shin-devigged closing line within noise on team-strength markets across 6 independent corpora. The one measured win is in-game conditioning (pregame prior fused with realized game state), labelled as calibration, not a betting edge.
• A reject graveyard and retraction log: 500+ candidate signals killed by the gate with reasons recorded; early headline numbers that did not survive adversarial re-audit are documented as retractions, not quietly deleted.
• A receipt system: every headline number on the public evidence surface carries its source artifact path, sha256, honesty label, and as-of date -- and a fail-closed linter blocks the bundle on edge-language or any retracted number.

Start here: docs/JOB_EVIDENCE_PACKET.md -- the adversarially-audited account of every claim, including the do-not-claim list.

A solo human directing a multi-model agent fleet: a frontier model decides, reviews, and adjudicates; mid-tier models execute in parallel against frozen contracts; small models scan. Hard guardrails live outside model discretion -- pre-commit hooks, fail-closed lint gates, append-only ledgers, binding invariant files. The patterns (and the failure modes) are written up honestly in agentic-fleet-playbook.

Stating this plainly because it decides whether the work is usable commercially. The ball detector in CourtVision fine-tunes Ultralytics YOLOv8, which is AGPL-3.0: fine for research and portfolio use, not fine inside a closed-source commercial product without a paid Ultralytics license. Shipping it commercially would mean retraining the detector on a permissively licensed backbone, which is a scoped swap, not a rewrite. The single-concept repos above carry no such constraint and are MIT.