Final-year CS. Open to a first role.

I take messy production data from raw rows to a decision a team can act on. And I publish what didn’t work.

Eight weeks on a real production search dataset: data contract, frozen baseline, model, honest validation, deployed paper. The write-up reports five archetypes instead of six, because I ran the check that knocked out the sixth.

Read the case study The paper Email me

What the rule hides

Fig. 01 k = 5

The hand-written rule puts 8,130 pages, 43% of the inventory, into one bucket and tells the team to monitor all of them. This is what’s actually inside it. The highlighted group of 735 carries a median 10,620 impressions at 0.42% CTR, and the rule says ignore them.

By the numbers

0 pages analysed

0 pseudonymized clients

0of 6 archetypes kept

0 leave-one-client-out mean ARI

Worst client 0.465. Directional, not established.

What you’ll find here

One case study, told properly.

What the problem was, what I decided, and what came of it, including the result I deleted. The case study. A short about. A way to reach me. Nothing else, because nothing else would help you decide.