Method _simple_pattern_routing

backend/server.py:506–552

📝 FALLBACK: Simple pattern-based routing (original logic).

(self, message: str, idx_ids: List[str])

            return self._simple_pattern_routing(query, [])

    def _simple_pattern_routing(self, message: str, idx_ids: List[str]) -> bool:
        """
        📝 FALLBACK: Simple pattern-based routing (original logic).
        """
        message_lower = message.lower()
        
        # Always use Direct LLM for greetings and casual conversation
        greeting_patterns = [
            'hello', 'hi', 'hey', 'greetings', 'good morning', 'good afternoon', 'good evening',
            'how are you', 'how do you do', 'nice to meet', 'pleasure to meet',
            'thanks', 'thank you', 'bye', 'goodbye', 'see you', 'talk to you later',
            'test', 'testing', 'check', 'ping', 'just saying', 'nevermind',
            'ok', 'okay', 'alright', 'got it', 'understood', 'i see'
        ]
        
        # Check for greeting patterns
        for pattern in greeting_patterns:
            if pattern in message_lower:
                return False  # Use Direct LLM for greetings
        
        # Keywords that strongly suggest document-related queries
        rag_indicators = [
            'document', 'doc', 'file', 'pdf', 'text', 'content', 'page',
            'according to', 'based on', 'mentioned', 'states', 'says',
            'what does', 'summarize', 'summary', 'analyze', 'analysis',
            'quote', 'citation', 'reference', 'source', 'evidence',
            'explain from', 'extract', 'find in', 'search for'
        ]
        
        # Check for strong RAG indicators
        for indicator in rag_indicators:
            if indicator in message_lower:
                return True
        
        # Question words + substantial length might benefit from RAG
        question_words = ['what', 'how', 'when', 'where', 'why', 'who', 'which']
        starts_with_question = any(message_lower.startswith(word) for word in question_words)
        
        if starts_with_question and len(message) > 40:
            return True
        
        # Very short messages - use direct LLM
        if len(message.strip()) < 20:
            return False
        
        # Default to Direct LLM unless there's clear indication of document query
        return False
    
    def _handle_direct_llm_query(self, session_id: str, message: str, session: dict):
        """