FIX THE PROCESS. FIND THE PEOPLE.

I Called It Dead on Friday. It Started Working on Saturday.

Fifteen days of OAuth hell — three Client IDs, four consents, a DocuSign agreement, and an "access is now live" email that did nothing. I gave up two days before the draft. Yahoo propagated overnight.

Two 401s that look identical are not the same failure. The diagnostic that broke it open was keeping the dead app as a control. Also: a float that only appears when you're losing, and a test that caught its own vacuity.

Using AI To Win My League

I've been running my fantasy auction draft off a static spreadsheet I made in July. This year I built an AI copilot that recomputes everything — inflation, opponent budgets, what to nominate — after every bid.

Real-time inflation math, per-opponent max bid tracking, a nomination engine that drains budgets strategically, and three years of behavioral data I'd been reading wrong. The tool told me what other people could afford. It mostly told me about me.

I Built a Real-Time Auction App Because Our Spreadsheet Had One Too Many Versions

Six guys, a $447 pot, and a Google Sheet that three people had edited simultaneously. There had to be a better way to run a Calcutta.

PlayCalcutta is a real-time Calcutta auction app — no accounts, no spreadsheet, just a 6-character code and a Firestore-backed leaderboard that updates live from ESPN while your friends argue about who overpaid for Rory McIlroy.

My Trading Bot Bet Against Trump Saying "Hottest." Trump Says "Hottest" Every 36 Hours.

I built an autonomous prediction market bot, deposited $350, and immediately lost money on a bet I should've won by a lot. Here's why, and what I learned fixing it.

A single Unicode curly-quote character bypassed my entire signal model and flipped a near-certain YES into a NO bet. I break down the four bugs that cost me money — and what the bot actually does when it's working right.

I Built a Recruiting Tool in a Weekend (And I'm Not an Engineer)

How 14 years of frustration and a conversation with an AI led to Sourcery — a multi-source talent sourcing agent that takes a job description and returns a ranked list of candidates in 15 seconds.

Sourcing was never the valuable part of recruiting. It was the tax you paid to get to the valuable part. Here's how I built a tool that compresses that overhead to near-zero — and what it actually means for the future of the role.

My Dream Talent Automation Stack

Screening 200+ applicants a week isn't scalable — not if you actually want to give each candidate a real experience. Here's the end-to-end automation workflow I'd build if I had full control of the stack.

I break down the full 4-stage workflow — from Tofu + Gem at the top of funnel, to an Ezra async AI screen, to Zapier routing Zoom transcripts into auto-filled scorecards, to a Greenhouse offer approval that fires DocuSign and a personalized video message automatically. The goal: protect the human moments by automating everything around them.