I treat math, code & a clean spreadsheet as the same kind of tool — a way to be precise about something fuzzy.
Things I built after hours.
A scratchpad — side projects, scripts, notebooks, half-finished ideas. Nothing here is part of a job. Most of it just had to exist.
redshift-warehouse-helpers
Tiny Python utilities for inspecting, profiling and snapshotting tables in Redshift.
Why I'd rather build the same query twice than ship a clever one.
# 14-day retention curve def retention(events): cohort = events.groupby( ["signup_d", "days"] )["uid"].nunique() return cohort.unstack().div( cohort.xs(0, level=1), axis=0 )
Open to a quiet collab on data or ML.
A year of small commits.
Most of these are evening sessions. Some are a single test passing.
A model is a story the data agreed to tell you, on a Tuesday.
— scribbled in a margin, 2025Dice coefficient on the brain-tumor U-Net side experiment (BraTS subset).
Currently re-reading "How to Take Smart Notes" — and ignoring most of it.
airflow-recipe-pack
DAG templates for daily ELT, retries, Slack alerts & cache warming.
Code in Place — Stanford
Six weeks of evening Python, plus a final mini-app. Kept the notebook, deleted the slack.
unet-mri-playground
Notebook companion to the TÜBİTAK research — U-Net variants, augmentation grids, Dice/IoU dashboards.