Generating the Answer

Question
+
Retrieved Context
+
Instructions
        ↓
       LLM
        ↓
Answer
+
Sources

The model should answer from the retrieved context, not invent information.