Hi everyone,
I am currently working on optimizing a RAG pipeline and researching GLiNER (Generalist and Lightweight Named Entity Recognition) integration.
I already understand the standard use case where we extract entities (like dates, locations, product_names) to create hard metadata filters (SQL-style filtering) before the vector search. This works well for narrowing the search space.
However, I am looking for architectural patterns and Python examples for more advanced implementations to improve retrieval accuracy. Specifically, I am interested in how to use extracted entities to influence the vector search directly, rather than just filtering it.