intent_parser.py

from typing import List, Dict, Tuple
import re

class IntentParser:
    """Parse user prompts to extract intent, roles, and constraints."""
    
    ROLE_KEYWORDS = {
        "backend": ["backend", "server", "api", "devops"],
        "frontend": ["frontend", "ui", "ux", "web"],
        "full-stack": ["full-stack", "full stack", "fullstack"],
        "web": ["web", "web dev", "web development"],
        "ai": ["ai", "artificial intelligence", "llm"],
        "ml": ["ml", "machine learning", "models"],
    }
    
    COMPANY_TYPE_KEYWORDS = {
        "startup": ["startup", "startups", "early stage"],
        "tech": ["tech", "technology", "saas"],
        "any": ["company", "companies"],
    }

    def parse_intent(self, user_prompt: str) -> Dict[str, any]:
        """
        Parse user prompt to extract:
        - Target roles
        - Company type
        - Location (if specified)
        - Any other constraints
        
        Returns:
            Dictionary with parsed intent
        """
        prompt_lower = user_prompt.lower()
        
        # Extract target roles
        target_roles = self._extract_roles(prompt_lower)
        
        # Extract company type
        company_type = self._extract_company_type(prompt_lower)
        
        # Extract location if mentioned
        location = self._extract_location(prompt_lower)
        
        # Determine if it's an outreach campaign
        is_outreach = self._is_outreach_intent(prompt_lower)
        
        return {
            "original_prompt": user_prompt,
            "target_roles": target_roles,
            "company_type": company_type,
            "location": location,
            "is_outreach": is_outreach,
            "confidence": self._calculate_confidence(target_roles, company_type)
        }

    def _extract_roles(self, prompt: str) -> List[str]:
        """Extract target roles from prompt."""
        found_roles = []
        
        for role, keywords in self.ROLE_KEYWORDS.items():
            for keyword in keywords:
                if keyword in prompt:
                    if role not in found_roles:
                        found_roles.append(role)
                    break
        
        return found_roles if found_roles else ["full-stack"]

    def _extract_company_type(self, prompt: str) -> str:
        """Extract company type preference."""
        for comp_type, keywords in self.COMPANY_TYPE_KEYWORDS.items():
            for keyword in keywords:
                if keyword in prompt:
                    return comp_type
        
        return "startup"

    def _extract_location(self, prompt: str) -> str:
        """Extract location if mentioned."""
        location_pattern = r'(san francisco|sf|nyc|new york|london|toronto|remote|anywhere)'
        match = re.search(location_pattern, prompt)
        return match.group(1) if match else "remote"

    def _is_outreach_intent(self, prompt: str) -> bool:
        """Determine if prompt is requesting outreach emails."""
        outreach_keywords = ["outreach", "email", "cold email", "reach out", "contact", "message"]
        return any(keyword in prompt for keyword in outreach_keywords)

    def _calculate_confidence(self, roles: List[str], company_type: str) -> float:
        """Calculate confidence score of intent parsing."""
        score = 0.5  # Base score
        score += min(len(roles) * 0.2, 0.3)  # More roles = higher confidence
        score += 0.2 if company_type != "startup" else 0  # Type specification helps
        return min(score, 1.0)

    def format_intent_summary(self, intent: Dict) -> str:
        """Format intent into readable summary."""
        roles = ", ".join(intent["target_roles"])
        return f"""
Intent Analysis:
- Target Roles: {roles}
- Company Type: {intent["company_type"]}
- Location: {intent["location"]}
- Is Outreach Campaign: {intent["is_outreach"]}
- Confidence: {intent["confidence"]:.0%}
"""