I suspect this is rate limiting or some error with embedding backfilling, cause when I try to parse search results I get invalid json from a json API.
I put a safeguard in place.
We can later search our AI logs for “Failed to search” to see if it is still happening @Falco and I guess push the result.body somewhere.
So the errors should be gone now, but if search is failing hallucinations will go up.
Had trouble reproing this so I suspect there is an edge case with our new embeddings.