Essays on production vector search, retrieval and the systems around them, published as The Dev Kernel.
01 / Current thread
Graph rot under writes and the hot/cold index pattern that answers it, redundancy in pure similarity ranking, and why recall@k measures your approximation rather than your product.
The Dev Kernel · Part 2 of 2
What changes past tens of millions of vectors: semantic sharding makes the index tractable and creates a router, and that router is a second layer of approximate search that fails silently.
The Dev Kernel · Part 1 of 2
02 / Earlier
These came out of what I was learning rather than what I had measured. They stay up: the dates are part of the record, and a publication that quietly removes its past reads worse than one that visibly changed direction.