Description
RelevanceTruthAndCompletenessEvaluator currently only considers the context present in the conversation history. When the LLM responses are based on grounding context that is supplied externally, the produced RTC scores are unreliable since the evaluation does not take this context into account. To fix this, RelevanceTruthAndCompletenessEvaluator should allow callers to (optionally) supply additional EvaluationContext and include this context as part of its evaluation prompt.