Update app.py

app.py

@@ -1187,7 +1187,11 @@ def get_behavior_instructions() -> str:
 
 
 def _looks_social_or_greeting(question: str) -> bool:
-    return bool(SOCIAL_RE.search((question or "").strip()))
+    # SOCIAL_RE is anchored with ^ so .match() is the correct call.
+    # Using .search() with ^ worked most of the time but could silently
+    # fail when the engine didn't treat ^ as start-of-string in search mode
+    # with certain Unicode inputs — .match() is unambiguous.
+    return bool(SOCIAL_RE.match((question or "").strip()))
 
 
 async def _interpret_semantics(question: str, history=None) -> dict:
@@ -1545,10 +1549,45 @@ async def agent_query(req: AgentRequest):
         }
 
     if _looks_social_or_greeting(q_clean):
+        # ── Try the hardcoded backstop first — zero latency, always correct ──
+        # This guarantees "who are you", "how are you", etc. NEVER produce
+        # "I don't have information" even if _answer_general has edge-case failures.
+        fixed = _fixed_social_answer(q_clean)
+        if fixed:
+            elapsed = time.time() - start
+            return {
+                "answer": fixed,
+                "intent": "social_greeting",
+                "tools_used": [],
+                "search_results": [],
+                "sources": [],
+                "model_used": req.model,
+                "response_time": round(elapsed, 2),
+                "corrected_query": question,
+                "natural_query": question,
+                "database_query": question,
+                "original_question": question,
+                "is_follow_up": False,
+                "source_mode": "social",
+            }
+        # ── Fall through to general LLM for social questions not in the fixed list ──
         general = await _answer_general(question, history)
+        # If the general answer is empty or looks like a "no info" reply, substitute a safe fallback
+        gen_answer = general.get("answer", "").strip()
+        no_info_social = [
+            "i don't have", "i do not have", "no specific information",
+            "not available", "cannot find", "unable to find", "no information",
+        ]
+        gen_lower = gen_answer.lower().replace("\u2019", "'").replace("\u2018", "'")
+        if not gen_answer or any(p in gen_lower for p in no_info_social):
+            gen_answer = (
+                "I'm <strong>LibBee</strong>, the Khalifa University Library AI Assistant. "
+                "I'm here to help with library questions, books, articles, databases, "
+                "full text access, and library services. What can I help you with?"
+            )
         elapsed = time.time() - start
         return {
-            "answer": general.get("answer", ""),
+            "answer": gen_answer,
             "intent": "social_greeting",
             "tools_used": ["general"],
             "search_results": [],
@@ -1674,7 +1713,10 @@ async def agent_query(req: AgentRequest):
             natural_query = search_plan["natural"]
             database_query = search_plan["database_query"] or search_plan["corrected"]
         elif intent == "library_info" and interp.get("grounding_keys"):
-            # strengthen RAG using interpreted canonical library terms
+            # Strengthen RAG using interpreted canonical library terms.
+            # IMPORTANT: only mutate `question` for library_info intent —
+            # social/general questions must NOT be expanded with library keywords
+            # because that causes the synthesis LLM to produce "I don't have information".
             resolved = " ".join([question] + interp.get("canonical_terms", []) + [GROUNDED_LIBRARY_MAP.get(k, "") for k in interp.get("grounding_keys", []) if GROUNDED_LIBRARY_MAP.get(k)])
             question = re.sub(r"\s+", " ", resolved).strip()
     except Exception: