Implementing case insensitive Vector Search filters

[Henry_Weller]

Hi @Ethan_kleinlugtenbelt! Thank you for the question, I can answer from the POV of the MongoDB product team.

We are looking at adding support for different types of lexical filters like regex and queryString to the $vectorSearch prefilter. In the meantime, you should be able to use a regex prefilter using the queryString operator as a filter within the knnBeta operator.

[
"$search": {
	"index": "vector_index_knnvector",
	"knnBeta": {
		"filter": {
			"compound": {
				"must": [
					{
						"queryString": {
							"query": "CAMBRIDGE AND test~2 NOT ALAN",
							"defaultPath": "descriptions"
						}
					},
				]
		},
		"vector": [...],
		"path": "seovoc:embeddings",
		"k": 2000
	}
},
{"$limit": 200}
]

Note that you will have to build a new index using the knnVector field type if you want to take this approach.

I will call out that while both of this operator and field type are deprecated, we would not EOL them before support for this and a few other types of filter in $vectorSearch are supported. Once that day comes you should be able to switch back to using $vectorSearch.

Hopefully this helps, and happy to answer any follow-ups you might have.