Fix: Humanize responses - increase tokens to 120, persist language override across turns

app/agents/orchestrator.py

@@ -417,6 +417,10 @@ class HoneypotOrchestrator:
                 if persona:
                      # Ensure the dict has the key so persona_engine knows which one it is
                      persona = {**persona, "selected_persona_key": existing_persona_key}
+                     # Apply stored language from session to preserve language override
+                     stored_lang = conversation.get("aggregated_intelligence", {}).get("request_language")
+                     if stored_lang:
+                         persona["language"] = stored_lang
                 self.logger.info(f"PERSONA LOCKED: Reusing {existing_persona_key}", session_id=conv_id)
             
             if not ctx.persona_locked:
@@ -445,11 +449,20 @@ class HoneypotOrchestrator:
         
         # [FIX] LANGUAGE OVERRIDE: Apply GUVI request language to persona
         # This ensures responses match the requested language (English/Hindi/Hinglish)
+        # Also persist language in session for subsequent turns with persona lock
         if request_language and persona:
             original_lang = persona.get("language", "hinglish")
             persona["language"] = request_language
             persona["original_language"] = original_lang  # Preserve for debugging
+            # Store in session for persistence
+            if ctx and hasattr(ctx, "session") and ctx.session:
+                ctx.session["request_language"] = request_language
             self.logger.info(f"Language override: {original_lang} -> {request_language}", session_id=conv_id)
+        elif ctx and hasattr(ctx, "session") and ctx.session:
+            # Restore language from session if not in current request
+            stored_lang = ctx.session.get("request_language")
+            if stored_lang and persona:
+                persona["language"] = stored_lang
         
         # [LIFECYCLE] Recalculate finalization state based on newly extracted intel
         # This ensures that if we just captured a UPI ID, we trigger XAI immediately.
@@ -637,6 +650,10 @@ class HoneypotOrchestrator:
             agg_intel = ctx.session.get("aggregated_intelligence", {})
             if agg_intel.get("metadata_agitation_reason"):
                  intelligence["metadata_agitation_reason"] = [agg_intel["metadata_agitation_reason"]]
+        
+        # 🔥 LANGUAGE PERSISTENCE: Store request language for persona lock on subsequent turns
+        if request_language:
+            intelligence["request_language"] = request_language
 
         await self.conversation_manager.update(
             conversation_id=conv_id,

app/agents/persona_engine.py

@@ -1022,7 +1022,7 @@ class PersonaEngine:
             prompt=formatted_prompt,
             role=ModelRole.FAST_CHAT,
             temperature=0.9,
-            max_tokens=60,
+            max_tokens=120,  # Increased for more natural responses
             context=context
         )