Founder-Engineer
For two years I built and ran ShareShark solo, a real-money, dual-currency sweepstakes prediction platform: the money ledger and ACH / KYC / AML compliance stack, a calibrated ML pricing engine, and a suite of AI risk agents. It ran in production, a year of free-to-play plus a real-money soft launch.
I came up through poker and +EV markets, so I think in probabilities, calibration, and survivable risk, and I pick up new domains fast. B.B.A. Finance, magna cum laude (UGA).
What I build
Money-correct systems: concurrency-safe ledgers, exactly-once payouts, ACH disbursement, KYC/AML, geofencing & fraud.
AI shipped in production: multi-agent systems, two-tier LLM routing, fail-closed serving, model output treated as untrusted.
Calibration-first ML pricing, Monte Carlo, copulas, Kelly sizing, and edge measured by closing-line value.
Case studies
An idempotent ACH payout lifecycle, KYC/AML gating, encrypted PII, tax reporting, and multi-signal geofencing (with a CGNAT/GPS rescue): a dozen third-party rails wired to fail closed.
Read the case study → Applied-AIA suite of cooperating Claude agents (two-tier Haiku/Sonnet routing, LLM output treated as untrusted, shadow-mode rollout) guarding the platform 24/7 against stale-price exploitation.
Read the case study →A LightGBM pricing engine (0.979 AUC / 0.002 ECE in backtest) with calibration-first evaluation and a model-free "delta-ceiling" bound.
Read the case study → Quant · MLTwice I bet that more-specialized models would price better; honest backtests said ship the single robust model. A study in experiment discipline.
Read the case study → Quant · MLMultiplying correlated probabilities overpays users on same-sector stacks. A Student-t copula Monte Carlo over a Ledoit-Wolf, regime-aware correlation matrix prices the joint probability, validated margin-positive across a six-year backtest.
Read the case study → Data Engineering · MLSix vendors, three timestamp formats, a million-plus API pulls, and point-in-time correctness throughout: backward merge_asof, expiry-only labels, and a purged-and-embargoed split that verifies its own honesty.
Selected projects
About
I'm a founder-engineer who builds production systems, from data pipelines and ML models to live serving and safety rails. Most of that came from building ShareShark solo, a real-money, dual-currency sweepstakes prediction platform (the dual-currency model is what keeps it a skill contest rather than gambling): a concurrency-safe money ledger, an ACH payout and KYC/AML/tax compliance stack, a calibrated LightGBM options-pricing engine, a suite of cooperating Claude-powered risk agents, and a cross-platform Vue/Capacitor client.
It ran in production through a year of free-to-play and a limited real-money soft launch; I paused the public real-money rollout at the payment-processor step after a closer read of the regulatory picture, and it stays live as free-to-play today. Alongside it I built calibrated sports-prediction models and a from-scratch quant risk toolkit. I treat model output as something to bound, not blindly trust, and I care most about systems that stay correct under real money and real users. Finance degree, pricing & valuation emphasis (UGA, magna cum laude).
Contact
Open to roles in regulated-fintech engineering, applied-AI / forward-deployed engineering, and quant / pricing & risk.
cooper@coopernorman.dev · linkedin.com/in/coopernorman1 · github.com/coopernorman
Download résumé (PDF) · See ShareShark live (shareshark.app), test login on request