AI builder & consultant · available for freelance work

I build agentic systems that run themselves.

I take a messy problem, build the quantitative pipeline, and wrap it in a system that runs itself. Fifteen years in data and statistics; since 2025, building and operating production AI systems, from problem to working prototype in hours.

See the work ↓ Contact me → Résumé (PDF)
15+years in data & stats
M.S.Statistics · Carnegie Mellon
hoursidea → prototype
solodesign → build → ship

Selected work

01

EV Blacksite ● live

Flagship · production SaaS · designed & built solo

A production, multi-service SaaS I designed, built, and shipped entirely on my own. Under the hood it's a betting-intelligence platform that pulls live odds from 80+ sportsbooks, computes each market's true price, and surfaces the best bets in real time. But the real story is the build: an AI chat assistant, a public API, an MCP server for agents, backtesting, and a full data pipeline on AWS, all from one person.

Next.jsExpressPython/ETLAWS ECSMCPPostgres+Athena
EV Blacksite live signal feed
screenshotLive dashboard
The platform in production
evblacksite.com ↗
02

Commission intelligence tool

Client build · internal business app · a regional auto-dealership group

A commission and compensation system I built for a client: per-person plans with bonuses, chargebacks and loans, a manager review-to-finalize approval workflow, 2FA and a full audit log, and an Anthropic + Gemini vision pipeline that parses sales screenshots straight into structured data.

Next.js 15TypeScriptSQLite/DrizzleLLM vision
Commission intelligence tool data import
screenshotCommission dashboard
Per-person plans, approvals, audit trail
03

EVB-Studio

My content pipeline · speeds up production

A pipeline I built for my own channel. It turns each day's results into most of a finished video (motion graphics, captions, and audio mastering) through an LLM-directed workflow, so the heavy editing is already done. I record the voiceover myself, then review, comment, and post by hand.

Pythonffmpegfaster-whisperLLMmotion graphics
EVB-Studio video: what's a half-run worth?
videoThe Daily Ledger
A recent episode. Watch the series on YouTube.

EV Blacksite on YouTube ↗

04

Automated betting execution ● runs all day

A product of EV Blacksite · autonomous trading system

An automated trading system that runs all day on EV Blacksite's own verified math and strategies. I added an MCP layer on top of the platform so the whole execution path (pricing, sizing, and placement) runs unattended on a fixed schedule, with fail-closed safety and self-correction. It's not an LLM guessing; it's back-tested, verified edges, executed automatically.

PythonMCPPlaywrightschedulerfail-closed safety
Automated betting execution, live status in Discord
screenshotLive status · Discord
Live status reported to Discord
05

The East Wind ● demo

A side project I build with my wife

An online multiplayer Taiwanese mahjong game my wife and I are building together in our spare time, meant to be played with friends over Zoom, Hangouts, or FaceTime. It's a playable demo: a 16-tile TypeScript engine, real-time multiplayer, a Claude Vision feature that reads a board screenshot into game state, AI coaching that helps new and experienced players alike, and an AI-generated soundtrack. Still far from finished.

TypeScriptReactVitestClaude VisionAI coachingAI music
The East Wind, Taiwanese mahjong with AI coaching
screenshotThe East Wind in play
16-tile Taiwanese mahjong, playable demo with AI coaching
theeastwind.gg ↗ (demo)

How I work

Every system I build is a loop that compounds.

  1. 01
    Identify the problem

    Watch the system in the wild and find where it leaks, stalls, or gets a decision wrong.

  2. 02
    Design a solution

    Model it, then pick the smallest change that actually moves the number.

  3. 03
    Let it execute

    Ship it and let it run unattended, with fail-closed safety and a graded record.

  4. 04
    Review the output

    Measure the result against reality, learn from it, and start the next pass sharper.

Each pass feeds the next, so the loop gets better every cycle, and that's the compounding. It's how the automated betting system runs and improves itself, on a schedule, without me babysitting it.

What I build for clients: autonomous agents & loops · quantitative data pipelines · AI-augmented internal tools.

About

Statistician by training (Carnegie Mellon), builder by disposition. Fifteen-plus years turning data into decisions, and since 2025 a full-time student of what AI makes possible.

I started in risk modeling at Agilex/Accenture, where a cargo-scoring model I led captured up to three times as many narcotics seizures as its predecessors. I automated client reporting at Orchard, then ran all reporting and business intelligence at Kabbage (now American Express), building a full-stack reporting service and real-time dashboards, and cutting one credit-line process from a full day to about fifteen minutes.

Then I stepped back for a sabbatical, to be a new dad and plan the next stage, while I kept consulting on data and never stopped building. In 2025 the AI wave clicked, and it became part of my daily life. I've built the systems and habits that let me build faster every week: think a problem all the way through, and have a working prototype in hours.

Off the clock

I build tools for my own obsessions.

StarCraft II analytics

Frame-level replay parsing into a time-series database, with an AI layer that coaches each new game against pro benchmarks.

Poker, quantified

An 86,000-hand PokerTracker analysis that found the true root-cause of losses: one opponent pool, proven at matched stakes.

Taiwanese mahjong

I love the game enough that my wife and I are building our own, The East Wind (above), a 16-tile engine with computer-vision board reading.

HouseFS

My home modeled as a filesystem (294 zones, 212 storage units), so everything has one canonical location.