About

I'm a builder, data product manager, and leader from the back of the room.

I work at the intersection of social impact and technology, building the kind of data infrastructure that moves capital and capability toward mission-driven work.

Right now I lead data products and partnerships at Candid, cofound Roost with Josie Moss, and author Valet: an agentic operating system with 25 production skills running my PM work daily inside Claude Code.

How I got here

I started my first "company" at six. I lived near a golf course and turned lost balls into renewed chances by selling them back to the Bogeymen. I learned about customer problems, customer preferences, and pricing. I was hooked.

Twenty years later I started my professional career through service: chairing a climate non-profit chapter, ushering refugees through resettlement, teaching English to primary school children. From there I worked in international policy and federal elections. Across all of it, social systems thinking kept turning into the same question: where's the leverage?

Technology was the answer. It still is. Most of my career has been in data products: analytics, pipelines, machine learning, and natural language search. Those are the mechanisms that close the gap between what stakeholders need and the insight, process, or clarity that gets them there.

Where I've worked

Education

What I'm optimizing for

Three tracks running in parallel, on purpose:

  1. Career. Staying sharp at the data PM craft, especially at scale and in mission-driven contexts. Open to builder-track roles in social-sector AI tooling and philanthropic infrastructure.
  2. Entrepreneurship. Roost is the active venture. Future ventures in the philanthropic-infrastructure space are on deck.
  3. Craft. Getting better at the agentic tooling layer: the orchestration, the memory systems, the places where AI genuinely augments judgment rather than simulates it. Valet is the testbed. The methodology essays are the output.

How I work

Why I work this way

I built Valet because the AI tools I was supposed to use assumed I had reliable internal scaffolding: that I'd remember to check, hold context across sessions, and catch what the model got wrong before it shipped. That assumption doesn't hold for me the way it might for other operators.

So I designed around the inversion. The system holds the structure. I hold the narrative. That's the principle the whole framework is built on, and it comes from a pretty specific place: organizational psychology, behavioral design, and a decade of watching smart people fail at execution not because they lacked intelligence but because their operating environment didn't support how they actually think.

The same principle runs through everything I build. Roost routes around expert critique because critique stalls people who already doubt themselves. Floodplain makes the trade-off visible instead of issuing a verdict. Observatory gets quieter when markets get loud. The design is always about which way to point the friction.

What I read / who I learn from

View CV · LinkedIn · Get in touch