Patrick McDowell
Investment Risk / Portfolio Construction / Front-Office Engineering
Looking back, I've been doing the same job my whole career. Only the titles changed. I find the friction people have quietly accepted as normal, and I design it away.
Today that means investment risk at a $1.4T global asset manager: working with Barra and Wellington's proprietary risk model across equity portfolios, and building the Python tools portfolio managers use every day. On nights and weekends, it means the projects below.
Turning analytics investors once tolerated into products they open every morning.
Investment Risk Analyst on Risk Strategy, Equity. I use Barra risk models alongside Wellington's proprietary internal risk model across the equity portfolios I cover, and work directly with portfolio managers on what they show. I also designed and backtested an optimized lower-risk equity strategy that helped win an institutional client pitch.
The firm's flagship equity risk platform. It started as a side idea, not a mandate. It's now opened daily by 100+ portfolio managers, risk managers, and senior leaders, and it collapsed six systems' worth of fragmented analytics into one workflow built around how investors actually think.
An internal AI platform built on the Model Context Protocol. Knowledge retrieval, dashboard generation, and reusable AI workflows that let non-developers build their own tools, turning a bottleneck (me) into a self-service capability.
The work I'm proudest of lives behind a wall at an asset manager, so you can't click it. The rest is below.
AI Product · Live
Who to draft, who to pick up, who to start — and a way to check whether the advice was any good.
It started because nobody in my league wanted to write the weekly recap. It became the thing I actually needed: a backend that pulls every season of my ESPN league into Postgres and answers what I used to guess at — who's worth a waiver claim, who to start on a light night, what to pay for a star. What I care about most is that I can check it: replay the recommender against 616 real decisions from last season and score it against what the league actually did. 22 managers across three leagues use it now, and it answers questions in Claude through an MCP server that can explain any number but can't make one up.
boxoutfantasy.com →
Client Work · In Implementation
Sons & DaughtersNational SocietyBy right of descent, a place that is yours.
A national lineage society for descendants of people enslaved in America brought me in as a consultant to rethink how AI could improve their technology and workflows, from archive search to registrar review. This is the working mockup we designed together, now being implemented.
sdusmp.patrickmcdowell.dev →
iOS · In Beta Prep
Do what matters first. Earn the rest.
The version of me who decides what matters isn't the version holding the phone at 9pm. So Earned takes the commitment I made while thinking clearly and enforces it later: the apps I'd rather not open stay shielded until the workout is done, and it asks HealthKit instead of taking my word for it. The rule I care about most is that a commitment can get harder but never easier once it's set — if you want out early, a friend gets the vote, not you. It runs on my phone today; TestFlight is waiting on Apple's review.
A lot of AI work right now is a great demo nobody uses. The best AI I've built disappears into how people work.
Read: How I think about AI →