hi guys, does LlamaIndex uses embeddings at the underlying level to send the context? I'm trying to figure out how LlamaIndex actually works.