https://vamsialla.com/notes.html Short write-ups on building software and AI that holds up: what worked, what didn't, and the numbers behind it. en-us Fri, 09 Oct 2026 12:00:00 +0000 https://vamsialla.com/notes/eval-said-96.html https://vamsialla.com/notes/eval-said-96.html Fri, 09 Oct 2026 12:00:00 +0000 Evals How a generous metric hid retrieval misses, and why I published the lower number.