I haven’t posted to in a while because I spent the last two weekends (and most days in between) building a multi-model sports analytics system with an LLM, part of my AAEL work and directly relevant to my dissertation.
Before I share, a note from the heart: I’m a Grand Canyon University grad and I know some will feel uneasy about sports modeling. I respect that. We can disagree in good faith and still care about excellence, integrity, and stewardship of our gifts.
Why I’m sharing this:
Sports data is just data.
It’s as open and structured as the real estate and mortgage data I use every day. If you can pipeline MLS or MERS data, you can pipeline play-by-play, tracking, and pricing feeds.
LLM-assisted modeling matters to my research.
I’m testing how a large language model can coordinate multiple statistical models (top-down priors, bottom-up simulations, player-level micro-projections) and then route them through an execution layer (risk sizing, vetoes, and calibration). That orchestration pattern applies far beyond sports.
Monte Carlo isn’t “casino math”: it’s industry math.
Randomized simulation under uncertainty is foundational across sectors:
Credit & mortgage risk: portfolio loss distributions, prepay/Default paths
Project finance & RE dev: schedule/cost risk and contingencies
Energy & reliability: outage risk, grid capacity planning
Pharma & clinical trials: power analyses and adaptive designs
Manufacturing & quality: yield variability and tolerance stacks
Supply chain: demand shocks, inventory and service-level tradeoffs
Insurance & reinsurance: catastrophe frequency/severity models
What we built (at a glance):
A calibrated game-flow simulator (GENESIS),
A top-down prior from power ratings (AXIOM),
Player-level projections (HELIOS),
A market-anchored ensemble to control variance, and
An OPS layer (position sizing, drawdown brakes, CLV governance).
It’s reproducible (seeded), auditable (hashes/logs), and price-aware (EV at current/alternate lines).
If this makes you cheer, awesome. If it makes you uneasy, I still appreciate you. My goal isn’t to promote gambling; it’s to advance transparent, responsible modeling that transfers to real-world risk and decision problems.
Happy to compare notes on Monte Carlo, LLM-assisted workflows, or how this architecture ports to real estate and mortgage credit.
Robert Foreman | Doctoral Student
DET | Central Michigan University
mobile 480-415-0783
email forem1r@cmich.edu