Your question is Evaluating RAG Reliability. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you evaluate the accuracy and reliability of retrieval-augmented generation (RAG) and vector-based context retrieval outputs?
Explain a practical evaluation approach for both retrieval and generated answers. Define the datasets, metrics, test cases, human review process, security checks, and release thresholds you would use. Address hallucination, stale or conflicting source content, irrelevant retrievals, prompt injection, access-control leakage, and production drift.