Preferred approach to filtering

[rcoundon]

We're evaluating a move from AWS Bedrock Knowledge base to LlamaParse and would like some guidance on the target structure with LlamaIndex.
In Bedrock KB we use a metadata filter expression to reduce the number of documents that are considered to answer a user query. This filtering is done on a single attribute, of which there are 20 or so values, and is provided alongside the query from the user.
Is there an equivalent with LlamaParse, or might it make more sense to create a separate 'tool' for each value of the filter?

EDIT: We could also use a separate index in the vector store and choose which to use at runtime.