backend/intent_detector.py

import os
import re
import warnings
from typing import Dict, Tuple

# Suppress deprecation warning for google.generativeai
warnings.filterwarnings("ignore", category=FutureWarning, module="google.generativeai")

import google.generativeai as genai


class IntentDetector:
    """Detects user intent using Gemini API with fallback patterns."""
    
    def __init__(self, api_key: str = None):
        """Initialize intent detector."""
        self.api_key = api_key or os.getenv("GEMINI_API_KEY")
        if self.api_key:
            genai.configure(api_key=self.api_key)
        
        self.model = genai.GenerativeModel("gemini-2.5-flash") if self.api_key else None
        
        # Fallback patterns for intent detection
        self.lead_keywords = ["interested", "pricing", "cost", "plan", "business", "solution", "need help", "help me", "problem", "solve", "budget"]
        self.support_keywords = ["issue", "problem", "broken", "error", "not working", "help", "support", "ticket", "complaint", "fix", "order"]
        self.general_keywords = ["hello", "hi", "what do you do", "services", "about", "company", "info", "tell me"]
    
    def detect_intent(self, user_message: str) -> Tuple[str, float]:
        """
        Detect user intent using Gemini API.
        
        Args:
            user_message: User's message
        
        Returns:
            Tuple of (intent, confidence)
            Intent: "lead", "support", "general", or "unknown"
        """
        if self.model:
            return self._detect_with_gemini(user_message)
        else:
            return self._detect_with_patterns(user_message)
    
    def _detect_with_gemini(self, user_message: str) -> Tuple[str, float]:
        """Detect intent using Gemini API."""
        try:
            prompt = f"""Analyze the user's message and classify their intent into ONE of these categories:
- "lead": They want to know about services, pricing, features, or have a problem to solve (sales/business inquiry)
- "support": They have a technical issue or need customer support help
- "general": They're asking general questions about the company or making small talk
- "unknown": The message doesn't fit any category

User message: "{user_message}"

Respond ONLY with the intent category name and a confidence score (0-1) in this format:
INTENT: [category]
CONFIDENCE: [score]"""
            
            response = self.model.generate_content(prompt, safety_settings=[
                {
                    "category": "HARM_CATEGORY_HARASSMENT",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_HATE_SPEECH",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
                    "threshold": "BLOCK_NONE",
                },
                {
                    "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
                    "threshold": "BLOCK_NONE",
                },
            ])
            
            text = response.text.strip()
            
            # Parse response
            intent = "unknown"
            confidence = 0.5
            
            for line in text.split('\n'):
                if "INTENT:" in line:
                    intent = line.split("INTENT:")[-1].strip().lower()
                    intent = intent.strip('"\'')
                elif "CONFIDENCE:" in line:
                    try:
                        confidence = float(line.split("CONFIDENCE:")[-1].strip())
                    except:
                        confidence = 0.5
            
            # Validate intent
            valid_intents = ["lead", "support", "general", "unknown"]
            if intent not in valid_intents:
                intent = "unknown"
            
            return intent, confidence
        
        except Exception as e:
            print(f"Error in Gemini detection: {e}")
            return self._detect_with_patterns(user_message)
    
    def _detect_with_patterns(self, user_message: str) -> Tuple[str, float]:
        """Fallback pattern-based detection."""
        message_lower = user_message.lower()
        
        lead_score = sum(1 for kw in self.lead_keywords if kw in message_lower)
        support_score = sum(1 for kw in self.support_keywords if kw in message_lower)
        general_score = sum(1 for kw in self.general_keywords if kw in message_lower)
        
        scores = {
            "lead": lead_score,
            "support": support_score,
            "general": general_score
        }
        
        max_score = max(scores.values())
        
        if max_score == 0:
            return "unknown", 0.3
        
        intent = max(scores, key=scores.get)
        confidence = min(max_score / 3, 1.0)
        
        return intent, confidence
    
    def extract_email(self, text: str) -> str:
        """Extract email from text."""
        pattern = r'([a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,})'
        match = re.search(pattern, text)
        return match.group(1) if match else None
    
    def extract_name(self, text: str) -> str:
        """Extract user name from text (simple heuristic)."""
        patterns = [
            r"my name is (\w+)",
            r"i'm (\w+)",
            r"i am (\w+)",
            r"call me (\w+)"
        ]
        
        text_lower = text.lower()
        for pattern in patterns:
            match = re.search(pattern, text_lower)
            if match:
                return match.group(1).capitalize()
        
        return None