Ragas evaluates retrieval-augmented generation pipelines at the component level, using LLM-based, reference-free assessment instead of expensive human-annotated ground-truth data. section 6.5.2
Its core metrics separate retrieval quality from generated-response quality:
- Context Precision checks whether the most relevant retrieved chunks appear near the top of the results. section 6.5.2
- Context Recall checks whether retrieved documents cover all relevant aspects of the query. section 6.5.2
- Faithfulness checks whether the generated response remains factually supported by the retrieved documents. section 6.5.2
- Answer Relevancy generates artificial questions from the answer and compares them with the original question. section 6.5.2
Comparing RAG configurations
Track the four component metrics separately to identify retrieval or generation weaknesses, then use the composite RAGAS score to compare configurations and monitor improvements over time. section 6.5.2 section 6.6
Figure 6.7 RAGAS Core Metric extensively covers LLM Grounding and Response (Source)
Use the figure to inspect how the core metrics cover retrieved context, LLM grounding, and response quality. section 6.5.2



