The assistant performs well on keyword queries and poorly on intent-heavy questions. How would you improve it?

Instruction: Describe how you would improve RAG when literal matches work but conceptual queries do not.

Context: Tests how the candidate diagnoses the problem, chooses the safest next step, and reasons through recovery. Describe how you would improve RAG when literal matches work but conceptual queries do not.

Official answer available

Preview the opening of the answer, then unlock the full walkthrough.

I would assume the system is matching words better than it is matching intent. Keyword queries succeed because lexical overlap does most of the work. Intent-heavy questions fail when the retriever does not bridge paraphrase, abstraction, or implied constraints well enough.

I would inspect the failure set for patterns....

Upgrade to view official answer

Related Questions