Lorem ipsum: your baseline model is lying to you politelyOn calibration, and why a 0.91 AUC can still ruin your quarter.
Blog
Notes on the practice of data science. New posts when I have something worth saying.
Dolor sit amet: shipping a feature store nobody asked forA postmortem in three acts, with receipts.
Consectetur: a defense of the humble SQL joinYou don't need a graph database. You need an index.
Adipiscing: notebooks are where reproducibility goes to dieRun-all is a lifestyle, not a guarantee.
Elit: the 80/20 of experiment designMost of the value is in the two questions you ask before randomizing.