Tag6 posts
May 12, 2026
I thought the model was failing. Most of the time, the harness around it was.
Read more →May 11, 2026
Fast retrieval found candidate sources. The next problem was deciding which sources should travel together.
Read more →May 10, 2026
Qdrant was not a bad tool. It was just the wrong scale for a small curated portfolio corpus.
Read more →May 9, 2026
My chat worked, but it felt broken. The fix started with measuring the wait users actually felt.
Read more →May 8, 2026
A normal portfolio is optimised for scanning. I wanted something optimised for asking.
Read more →January 10, 2026
MoSCoW taught me how to label priorities. It didn’t teach me how to choose.
Read more →