When you’re creating your embedding you can store keywords from the content (using an LLM) in the metadata of each chunk which would positively increase the rel

When you’re creating your embedding you can store keywords from the content (using an LLM) in the metadata of each chunk which would positively increase the relevancy of results turned from the retrieval. LlamaIndex does this out of the box.