Using an llm to improve semantic search

For my website, I want to implement a search with complicated natural language queries like “Where can I find content for young, disabled people who are looking for education support, as well as content for IT support for old people?”.
A hybrid or semantic search won’t work in that case, as the query needs to be understood and probably translated into different sub queries. I was looking into llamaindex but all guides I can find are about RAG and generating new text BASED on the found content.
How can I use an llm to only understand the query input and then return the most relevant website pages saved in weaviate? Basically an improved semantic search

This concept is somewhat related to Generative Feedback Loops.

You can use the LLM to parse the question, query the database, and keep looping until it found an answer that satisfies the original query.

Thanks! So in practice instead of near_text search it’s really just a prompt like “return the most relevant id’s for this user query”?