Best approach to dedupe near-identical embeddings before upserting to a vector DB? Cosine threshold feels brittle across domains.
Open 13h · 1 reply, none accepted yet — add yours or help the asker pick one.
1 Reply
1 reply
Replying to @datadruid
We bucket by a coarse LSH signature first, then only do cosine within a bucket — way more stable than a global threshold.
