Create intent_classifier.py

services/intent_classifier.py

@@ -0,0 +1,217 @@
+"""
+Intent Classification Service
+Determines user intent to make the chatbot behave naturally
+"""
+
+import re
+from typing import Dict, Any
+
+
+class IntentClassifier:
+    """Classifies user intents to determine appropriate bot response"""
+    
+    # Keywords that indicate pipeline creation intent
+    PIPELINE_KEYWORDS = [
+        "extract", "summarize", "translate", "classify", "detect", 
+        "analyze", "process", "generate", "create pipeline", "build pipeline",
+        "run", "execute", "perform", "do", "get", "find", "identify",
+        "table", "text", "image", "signature", "stamp", "ner", "entities"
+    ]
+    
+    # Casual chat patterns
+    CASUAL_PATTERNS = [
+        r"^(hi|hello|hey|greetings|good morning|good afternoon|good evening)",
+        r"^(how are you|what's up|wassup)",
+        r"^(thanks|thank you|appreciate)",
+        r"^(bye|goodbye|see you|later)",
+        r"^(ok|okay|cool|nice|great|awesome)",
+        r"^(what can you do|what do you do|help|capabilities)",
+        r"^(who are you|what are you)"
+    ]
+    
+    # Question patterns that need informational response
+    QUESTION_PATTERNS = [
+        r"^(what|how|why|when|where|who|can you|do you|are you|is it)",
+        r"(help|explain|tell me|show me)"
+    ]
+    
+    @staticmethod
+    def classify_intent(user_message: str) -> Dict[str, Any]:
+        """
+        Classify user intent from their message
+        
+        Returns:
+            {
+                "intent": "casual_chat" | "question" | "pipeline_request" | "approval" | "rejection",
+                "confidence": float (0-1),
+                "requires_pipeline": bool,
+                "suggested_response_type": "friendly" | "informational" | "pipeline_generation"
+            }
+        """
+        message = user_message.strip().lower()
+        
+        # Empty message
+        if not message:
+            return {
+                "intent": "casual_chat",
+                "confidence": 1.0,
+                "requires_pipeline": False,
+                "suggested_response_type": "friendly"
+            }
+        
+        # Approval/Rejection patterns (for pipeline confirmation)
+        if message in ["approve", "yes", "y", "ok", "okay", "proceed", "go ahead", "do it"]:
+            return {
+                "intent": "approval",
+                "confidence": 1.0,
+                "requires_pipeline": False,
+                "suggested_response_type": "execute_pipeline"
+            }
+        
+        if message in ["reject", "no", "n", "cancel", "stop", "don't"]:
+            return {
+                "intent": "rejection",
+                "confidence": 1.0,
+                "requires_pipeline": False,
+                "suggested_response_type": "friendly"
+            }
+        
+        # Casual chat patterns
+        for pattern in IntentClassifier.CASUAL_PATTERNS:
+            if re.search(pattern, message, re.IGNORECASE):
+                return {
+                    "intent": "casual_chat",
+                    "confidence": 0.9,
+                    "requires_pipeline": False,
+                    "suggested_response_type": "friendly"
+                }
+        
+        # Question patterns (informational)
+        is_question = False
+        for pattern in IntentClassifier.QUESTION_PATTERNS:
+            if re.search(pattern, message, re.IGNORECASE):
+                is_question = True
+                break
+        
+        # Check for pipeline keywords
+        pipeline_keyword_count = sum(
+            1 for keyword in IntentClassifier.PIPELINE_KEYWORDS 
+            if keyword in message
+        )
+        
+        # If has pipeline keywords, it's likely a pipeline request
+        if pipeline_keyword_count > 0:
+            return {
+                "intent": "pipeline_request",
+                "confidence": min(0.6 + (pipeline_keyword_count * 0.1), 1.0),
+                "requires_pipeline": True,
+                "suggested_response_type": "pipeline_generation",
+                "keyword_matches": pipeline_keyword_count
+            }
+        
+        # Questions without pipeline keywords
+        if is_question:
+            return {
+                "intent": "question",
+                "confidence": 0.8,
+                "requires_pipeline": False,
+                "suggested_response_type": "informational"
+            }
+        
+        # Default: treat as casual if short, otherwise might be pipeline request
+        if len(message.split()) < 3:
+            return {
+                "intent": "casual_chat",
+                "confidence": 0.6,
+                "requires_pipeline": False,
+                "suggested_response_type": "friendly"
+            }
+        
+        # Longer messages without clear intent - ask for clarification
+        return {
+            "intent": "unclear",
+            "confidence": 0.4,
+            "requires_pipeline": False,
+            "suggested_response_type": "clarification"
+        }
+    
+    @staticmethod
+    def get_friendly_response(intent: str, user_message: str = "") -> str:
+        """Generate friendly chatbot responses for non-pipeline intents"""
+        
+        message_lower = user_message.lower().strip()
+        
+        # Greetings
+        if re.search(r"^(hi|hello|hey)", message_lower):
+            return "Hello! 👋 I'm MasterLLM, your AI document processing assistant. Upload a document and tell me what you'd like to do with it!"
+        
+        # How are you
+        if "how are you" in message_lower:
+            return "I'm doing great, thank you! 🤖 Ready to help you process documents. Upload a file to get started!"
+        
+        # Thanks
+        if re.search(r"^(thanks|thank you)", message_lower):
+            return "You're welcome! 😊 Let me know if you need anything else!"
+        
+        # Goodbye
+        if re.search(r"^(bye|goodbye)", message_lower):
+            return "Goodbye! 👋 Feel free to come back anytime you need document processing help!"
+        
+        # Capabilities question
+        if "what can you do" in message_lower or "capabilities" in message_lower:
+            return """I can help you with various document processing tasks:
+
+📄 **Text Operations:**
+- Extract text from PDFs and images
+- Summarize documents
+- Translate to different languages
+- Classify text content
+- Extract named entities (NER)
+
+📊 **Table Operations:**
+- Extract tables from documents
+- Analyze tabular data
+
+🖼️ **Image Operations:**
+- Describe images
+- Detect signatures
+- Detect stamps
+
+🔧 **How to use:**
+1. Upload a document (PDF or image)
+2. Tell me what you want to do (e.g., "extract text and summarize")
+3. I'll create a pipeline for you to approve
+4. Watch the magic happen! ✨"""
+        
+        # Who are you
+        if "who are you" in message_lower or "what are you" in message_lower:
+            return "I'm MasterLLM 🤖, an AI-powered document processing orchestrator. I use advanced AI models (Bedrock Claude & Google Gemini) to understand your requests and automatically create processing pipelines for your documents!"
+        
+        # Help
+        if message_lower in ["help", "?"] or "help me" in message_lower:
+            return """Here's how to use me:
+
+1️⃣ **Upload Document**: Click the upload button and select a PDF or image
+2️⃣ **Describe Task**: Tell me what you want (e.g., "extract text from pages 1-5 and summarize")
+3️⃣ **Review Pipeline**: I'll show you the processing plan
+4️⃣ **Approve**: Type 'approve' to execute or 'reject' to cancel
+5️⃣ **Get Results**: Watch real-time progress and get your results!
+
+**Example requests:**
+- "extract text and summarize"
+- "get tables from pages 2-4"
+- "translate to Spanish"
+- "detect signatures and stamps"
+
+Need anything else?"""
+        
+        # Unclear intent
+        if intent == "unclear":
+            return "I'm not sure what you'd like me to do. Could you please:\n- Upload a document first, or\n- Tell me what processing task you need (e.g., 'extract text', 'summarize', 'translate')\n\nType 'help' to see what I can do!"
+        
+        # Default friendly response
+        return "I'm here to help! Upload a document and tell me what you'd like to do with it. Type 'help' if you need examples! 😊"
+
+
+# Singleton instance
+intent_classifier = IntentClassifier()