Update src/qa.py

src/qa.py

@@ -278,9 +278,13 @@ def truncate_context(context_text: str, max_tokens: int = 100000, model: str = "
         return truncated
     return context_text
 
+# ==========================================================
+# 8️⃣ Answer Generation (English Only + Token-Safe + Smart Fallback)
+# ==========================================================
 def generate_answer(query: str, retrieved_chunks: list, reasoning_mode: bool = False):
     """
     Generates an English answer using GPT-4o (SAP Gen AI Hub proxy).
+    Handles both strict and reasoning modes with smart fallback guidance.
     """
     if not retrieved_chunks:
         return "Sorry, I couldn’t find relevant information in the document."
@@ -312,7 +316,24 @@ def generate_answer(query: str, retrieved_chunks: list, reasoning_mode: bool = F
 
     try:
         response = chat_llm_local.invoke(messages)
-        return response.content.strip()
+        output = response.content.strip()
+
+        # 🔍 Smart fallback substitution
+        if "I don't know based on the provided document" in output:
+            if reasoning_mode:
+                output = (
+                    "I couldn’t infer enough from the context. "
+                    "Try rephrasing your question for a clearer reasoning path."
+                )
+            else:
+                output = (
+                    "I couldn’t find a clear answer in this document. "
+                    "You can try rephrasing the query or switch to Extended Mode "
+                    "(Document + General) for a broader explanation."
+                )
+
+        return output
+
     except Exception as e:
         print(f"⚠️ GPT-4o generation failed: {e}")
         return "⚠️ Error: Could not generate an answer."