I’m currently at Canva, tinkering with RL, agents, and ad generation.

I was previously CTO and co-founder at MangoAI, where we worked on applying reinforcement learning to generative video ads. We also spent a lot of time thinking about how to bring users along on the RL journey. MangoAI was acquired by Canva in early 2026.

I used to do various things with generative AI, machine learning, and causal inference at Roblox, Netflix, IBM Research, Stanford (EE), and MIT (RLE).

I generally enjoy working at the intersection of data, algorithms, and human creativity / content.

I once had a brief side hustle as a talking head about various topics around AI and ML (TED radio hour, IMDB).

I’m deeply passionate about film and theater (mostly on the consumption side these days).

I like dogs and dark chocolate.

Sporadic Posts

Evals, R1-Zero, and RL

Ephemeral code (and yes, agents)

Evals, tests, and LLM applications

Two quick takeaways from Karpathy’s talk

Trained evaluators and multi-subject finetuning

Real estate hypotheses

https://www.youtube.com/watch?v=zgTRPjHVrFE