fix(guvi): Resolve Handshake Loop - Strict Schema (Data vs Reply) & Data Population

app/agents/orchestrator.py

@@ -202,6 +202,7 @@ class HoneypotOrchestrator:
         scammer_behavior = None
         escalation_rec = {}
         is_fast_path = False
+        is_scammer_repeating = False # [BUG FIX] Initialize early for Fast Path safety
 
         # [SCORING] Role-based logic: If sender is 'user', treat as non-scam or testing turn
         if sender_role == "user":

app/config.py

@@ -47,12 +47,13 @@ class Settings(BaseSettings):
     GROQ_MODEL: str = "llama-3.3-70b-versatile"
     
     # Per-task model routing (Production Grade)
-    GROQ_FAST_MODEL: str = "meta-llama/llama-4-scout-17b-16e-instruct" # 10M Context ✅, 30K TPM
-    GROQ_SMART_MODEL: str = "llama-3.3-70b-versatile"     # High RPM Reasoning, JSON Schema ✅
-    GROQ_NATURAL_MODEL: str = "meta-llama/llama-4-scout-17b-16e-instruct" # 10M Context ✅
-    GROQ_STRUCTURED_MODEL: str = "llama-3.3-70b-versatile" # Official JSON Schema Support
-    GROQ_SAFETY_MODEL: str = "meta-llama/llama-guard-4-12b" # 14.4K RPD, Production
-    GROQ_SAFEGUARD_MODEL: str = "openai/gpt-oss-safeguard-20b"  # Cache ✅, Policy-safe
+    # Per-task model routing (Production Grade)
+    GROQ_FAST_MODEL: str = "llama3-8b-8192" # Blazing fast (~300 tokens/sec)
+    GROQ_SMART_MODEL: str = "llama3-70b-8192" # Balanced reasoning
+    GROQ_NATURAL_MODEL: str = "llama3-8b-8192" 
+    GROQ_STRUCTURED_MODEL: str = "llama3-70b-8192" # Reliable JSON
+    GROQ_SAFETY_MODEL: str = "llama-guard-3-8b" 
+    GROQ_SAFEGUARD_MODEL: str = "llama-guard-3-8b"
 
     OPENROUTER_MODEL: str = "meta-llama/llama-3.1-70b-instruct"
     

app/utils/guvi_handler.py

@@ -183,15 +183,34 @@ class GUVIHandler:
             is_finalizing_turn = (db_history_len + 2) >= 2
             
             logger.debug("🔥 Orchestrator reached") # [DEBUG] Verify flow
-            result = await orchestrator.process_message(
-                message=scammer_text,
-                sender_id=sender,  # [SCORING] Align with forensic audit recommendation
-                sender_role=sender,  # [BUG FIX] Restore role for fail-safe engagement
-                conversation_id=session_id,
-                auto_report=True,
-                client_ip=client_ip,
-                should_finalize=is_finalizing_turn # [RESTORED] Critical for callback trigger
-            )
+            try:
+                # [LATENCY] Strict 25s Timeout to satisfy GUVI's 30s limit
+                # If LLM hangs, we abort and return fallback immediately
+                result = await asyncio.wait_for(
+                    orchestrator.process_message(
+                        message=scammer_text,
+                        sender_id=sender,  # [SCORING] Align with forensic audit recommendation
+                        sender_role=sender,  # [BUG FIX] Restore role for fail-safe engagement
+                        conversation_id=session_id,
+                        auto_report=True,
+                        client_ip=client_ip,
+                        should_finalize=is_finalizing_turn # [RESTORED] Critical for callback trigger
+                    ),
+                    timeout=25.0
+                )
+            except asyncio.TimeoutError:
+                logger.error(f"⏱️ DATA TIMEOUT ({session_id}): Orchestrator took >25s. Forcing fallback.")
+                # Construct a minimal valid 'result' to allow fall-through to standard response builder
+                result = {
+                    "status": "partial_success",
+                    "is_scam": False, # Fail open (continue)
+                    "threat_level": "MEDIUM", 
+                    "honeypot_response": {"message": "Hello? Thoda network slow hai mera.", "persona": "fallback"},
+                    "conversation": {"message_count": db_history_len + 1},
+                    "aggregated_intelligence": conv.get("aggregated_intelligence", {}),
+                    "confidence": 0.0,
+                    "agent_notes": "Latency Timeout - Fallback Triggered"
+                }
             
             # [SCORING] Accurate message counting (Forensic Fix)
             # Orchestrator returns 'message_count', history list is not guaranteed in result
@@ -384,8 +403,8 @@ class GUVIHandler:
                    bankAccounts=[], upiIds=[], phishingLinks=[], phoneNumbers=[], suspiciousKeywords=[]
                 ),
                 agentNotes=f"System Failover Triggered: {safe_error}",
-                reply="System under high load. Please retry.",
-                honeypotResponse="System under high load."
+                reply="Hello? Awaaz nahi aa rahi... network issue lag raha hai.",
+                honeypotResponse="Hello? Awaaz nahi aa rahi... network issue lag raha hai."
             )