Hybrid search was raggio’s widest loss in the first benchmark: a median of 104 ms against Weaviate’s 36, and a p99 near 600. The first post blamed …
The first post ended its ingest section with a line I was rather proud of: “batching those syncs is a known lever I haven’t pulled, because …
Part 1 was the loop. This part is about the first thing the loop has fixed.
Sooner or later every RAG system sends you the same bill, and it’s written in gigabytes of RAM. The model wants memory for its weights and its KV cache. …