Analytics engineering and product analytics leader with 10+ years across games, social platforms, insurance, and finance — currently a Senior Data Scientist at Epic Games on Fortnite Discovery. Data Investors is the nights-and-weekends proof of how I like to build: end to end, tested, and honest about its own performance.
At Epic Games I'm the sole data scientist for Fortnite Discover — the surface where millions of daily players find what to play next. I own its analytics data models end to end: a dbt-on-Databricks architecture that turns raw event telemetry into tested source, staging, fact, and aggregate layers, with incremental rebuild macros, daily data-quality checks, and Airflow orchestration. The habit transfers: this site runs on the same layered, tested pattern, one Postgres warehouse instead of a lakehouse.
Trusted foundations also mean hunting the silent failures — at Epic I uncovered production defects that had been NULLing key columns for weeks and codified the telemetry caveats that keep biased metrics out of product decisions.
Numbers only matter when they move a decision. At Rec Room, my user-level LTV models showed non-mobile players were worth 3–8x more than mobile — leadership pivoted the platform strategy. My geo-lift testing exposed attribution gaps understating ad ROAS by ~4x, and the CEO restarted paid acquisition. At Meta's Horizon Worlds, showing that a day-one social connection made users 5.6x more likely to retain through day 28 helped redefine the product's target metrics.
The same discipline runs through my experimentation work: sample-ratio-mismatch checks, A/A gates, exposure-semantics validation — the boring machinery that keeps false lift signals out of ship decisions.
From board-level financial models that helped raise $350M+ at Root Insurance, to executive dashboards and custom analytical tooling at Epic, Rec Room, and Meta (Tableau, Hex, Sigma, Plotly-style web tooling), I build the artifact the audience needs — not the one the tool defaults to. This site is exhibit A: every chart on the preview pages was designed to make a probability, its base rate, and its track record readable at a glance.
At Epic I built an AI-assisted analytics system — versioned domain skills, 37 canonical metric contracts, multi-agent investigation workflows, and evaluation harnesses that regression-test the system itself. This platform is the personal-scale version of the same conviction: its 2026 rebuild — scrapers, warehouse, models, and this site — was executed in days, not months, using agentic AI workflows with 200+ automated tests as the safety net. Iteration beats perfection, provided the tests are the ones deciding.
Every model answers the same auditable question — how likely is a barrier hit within the window? — and publishes the probability, not a story.
Publish-dated joins, walk-forward retraining, out-of-time grading. The backtest can't cheat, so the live numbers get to mean something.
Scrapers, warehouse, models, and this site are one person's stack — which is exactly why every number on it can be traced to its source.