Behind the notes.

I did my PhD in applied mathematics and statistics at Stony Brook, on how to use heavy-tailed distributions in portfolio allocation. Since then I have bounced between finance and data: risk and portfolio research, models for mobile location data, a recommendation algorithm for books, and now risk models again.

The thread through all of it is the same annoying fact: real data has fat tails, and the textbook shortcuts break exactly when it matters.

normix is my attempt to make that less painful. It's a JAX package for normal variance mixture distributions, with generalized hyperbolic as the main case, and it covers fitting them to data and computing risk measures under them. Longer write-ups are in the math notes, shorter ones on the blog.

I also spend a lot of time pair-programming with AI agents, which is probably worth a post once I have figured out what I actually think about it. My Cursor profile has the receipts.