Evaluations

Groundedness, relevance, recommendation quality, policy compliance, tool selection, hallucination, action safety, structured output and latency.