Home / Blog

Blog

Notes from building data work: what broke, how I found it, what I changed and what it cost. Written for analysts, engineers and anyone hiring them.

CI/CDCareerCostData analyticsData leakageGitHub ActionsKaggleMLOpsMachine learningValidation

I used a month of GitHub Actions minutes in one evening

2,391 of 2,000 free minutes, gone in about five hours. The billing detail I missed, and the guardrails I now put on every workflow.

GitHub ActionsCI/CDMLOpsCost

When the validation set lies: a leak-free hold-out for CASMI 2026

My offline score said 0.930. The leaderboard dropped from 0.415 to 0.371. How I built a 300-molecule hold-out that finally agreed with the leaderboard.

KaggleValidationMachine learningData leakage

Why I'm writing in public

Who I am, what this blog will cover and why every post comes with the numbers and the parts that didn't work.

CareerData analytics