/now/

6 months on the team that runs stripe's apache spark infrastructure. worked on autotuning spark configs automatically (so engineers don't have to fiddle with executor memory by hand), cpu oversubscription to squeeze more jobs per machine, and making the spark history server actually reliable. joining full-time in july on the same team.

spark workflow for syncing mongodb → elasticsearch at scale. that one saved the company north of $600k/year in infra costs, which was a fun thing to find out.

self-supervised privacy preservation for action recognition. how do you train a model on video without leaking who's in the video. cool problem, very cold country.