About

I'm a tinkerer. As a kid that meant caring about the Game Boy as much as whatever was in it, and taking apart anything with screws. That hasn't really changed — I still can't use something well made without wanting to know how it was put together.

So the history is a generalist's. I went to university for computer science and hated it, switched to statistics, and found my way into machine learning through the back door. Then a family business sold and I spent two years running money across public markets, private equity, venture and property. Now I build software end to end.

How I got here

Four stages, and the thing at the end of each one is what pushed me into the next.

  1. StatisticsI went to university for computer science and hated it.Toronto · 2016–2020

    What I hated was the coding. What I loved was the math underneath it, which nobody tells you is the same subject wearing different clothes.

    So I switched to statistics, and then found the half-space between the two in statistical learning. That was the door into ML and I went through it.

  2. Machine learningI started out making graphs for the business team.Fundmore · 2021–2023

    I ended up building a new ML product, Fundmore IQ, doing document extraction for mortgage origination.

    The three things I actually learned there had nothing to do with modelling. How to use git properly. How much of “data work” is just manual work. And that explaining a technical thing to a non-technical person is its own craft — one I was bad at, and then less bad at.

  3. A break from techThen I left, because my family needed me more than a codebase did.2023–2024

    A business had just sold and somebody had to build whatever came after it. So I did: entity structure, mandate, allocation, across public markets, private equity, venture and property. Some of that year I spent on a construction site with a nail gun.

    It's a strange line to have in your history and it's the most useful thing I've done. I learned more about finance than a course would have taught me, and I learned what an operator's day actually looks like, which is the reason I now build for small operators instead of at them.

  4. ProductAnd then Claude Code showed up.2024–now

    I build end to end now. I finally don't need to be an expert in React to give my models a frontend.

    Which means the question I actually care about — does this work for the person using it — stopped depending on which half of the stack I happened to be better at.

What came out of all that is a preference for systems that refuse rather than guess. Confidence per field, provenance on every number, and a rule that sends the uncertain cases to a person. It shows up in almost everything I build, because the failure I find hardest to forgive is software that's confidently wrong.

I also build things from scratch to understand them. A GPT by hand in MLX instead of importing one. An eval harness because I couldn't find a straight answer to a question I had. Most of it isn't novel and that's fine — the point is that afterwards I know how it works.

Outside of that: soccer once a week, hackathons for the deadline rather than the prize, reading widely and taking terrible notes on purpose, and a knowledge vault my agents read and write to.

You can see what I've built on Projects, what I'm thinking about in my writing, or reach me at [email protected].