Understanding Pgvector's HNSW Index Storage in Postgres
https://lantern.dev/blog/pgvector-storageVector searching had strange quirks where searching for "cat" would return mostly a lot of paragraphs unrelated to the word. I was using 3072 length for OAI text-embedding-3-large. Each entry was roughly 1-2 paragraphs. For my recent project, I found that PGroonga was more reliable for full text document lookup (with some fuzzy matching support).
loading story #41453217
loading story #41455077
loading story #41454616
loading story #41456007
loading story #41454427
loading story #41456085
loading story #41454035