
context engineering, - only include vector hits that are x distance?

AI in the middle - make the ai generate the string for vector search

instruction tuned embeddings?

entity chunking & re-ranking

bredth vs depth = separate workflows

examples into prompts & better prompts

common model attributes - temp & top-k

QA specific embedding models? 



Evaluation metrics, how good is it doing? 
rate my response!? 