LLM Evaluation (RAGAS / G-Eval) & Safety Guardrails
Measure Faithfulness, Context Precision, Answer Relevance, and enforce input/output security guardrails against prompt injections.
LLM Evaluation Metrics & Guardrails
### RAG Triad Evaluation Metrics
- **Faithfulness**: Measures whether answer facts are strictly grounded in retrieved context (detects hallucinations).
- **Answer Relevance**: Measures how directly the generated answer addresses user query goals.
- **Context Precision**: Measures the signal-to-noise ratio of retrieved vector passages.
