"""
Generate management talking points for investor relations.
Uses free HuggingFace models with rule-based fallback.
"""
import re
from transformers import pipeline

_generator = None


def get_generator(model_name="google/flan-t5-base"):
    """Load text2text generation pipeline with fallback."""
    global _generator
    if _generator is None:
        try:
            _generator = pipeline(
                "text2text-generation",
                model=model_name,
                device=-1  # CPU only for free deployment
            )
        except Exception:
            _generator = None
    return _generator


def rule_based_talking_points(content, company_name, memo_style=False):
    """Fallback: Generate structured talking points using rules."""
    lines = [l.strip() for l in content.split('\n') if l.strip()]
    
    if memo_style:
        # Memo format
        talking_points = ["### Executive Summary\n"]
        talking_points.append(f"This memo outlines key strategic considerations and talking points for {company_name}.\n")
        
        talking_points.append("### Key Points\n")
        for i, line in enumerate(lines[:8], 1):
            if line and not line.startswith('#'):
                talking_points.append(f"{i}. {line}")
        
        talking_points.append("\n### Strategic Priorities\n")
        
        # Extract action-oriented items
        action_words = ['invest', 'launch', 'expand', 'acquire', 'focus', 'pursue', 'grow']
        for line in lines:
            line_lower = line.lower()
            if any(word in line_lower for word in action_words):
                talking_points.append(f"- {line}")
        
        talking_points.append("\n### Investor Considerations\n")
        talking_points.append("- Emphasize long-term strategic value creation")
        talking_points.append("- Highlight competitive advantages and market positioning")
        talking_points.append("- Address key risks and mitigation strategies")
        
    else:
        # Earnings call format
        talking_points = ["### Opening Remarks\n"]
        talking_points.append(f"Thank you for joining {company_name}'s earnings call today. I'd like to highlight our strong performance and strategic progress.\n")
        
        talking_points.append("### Performance Highlights\n")
        
        # Extract positive metrics
        positive_lines = []
        for line in lines:
            line_lower = line.lower()
            if any(word in line_lower for word in ['growth', 'beat', 'up', 'increase', 'strong', 'success', 'achievement', 'launch']):
                positive_lines.append(line)
        
        for line in positive_lines[:5]:
            talking_points.append(f"- {line}")
        
        talking_points.append("\n### Strategic Initiatives\n")
        talking_points.append("- Continue investing in innovation and product development")
        talking_points.append("- Expand market presence and customer base")
        talking_points.append("- Drive operational efficiency and margin improvement")
        talking_points.append("- Pursue strategic opportunities for growth")
        
        # Extract challenges and address them
        challenge_lines = []
        for line in lines:
            line_lower = line.lower()
            if any(word in line_lower for word in ['challenge', 'delay', 'headwind', 'pressure']):
                challenge_lines.append(line)
        
        if challenge_lines:
            talking_points.append("\n### Addressing Challenges\n")
            for line in challenge_lines[:3]:
                talking_points.append(f"- {line}")
                talking_points.append(f"  - *Mitigation*: Proactive management and contingency planning in place")
        
        talking_points.append("\n### Looking Forward\n")
        
        # Extract guidance
        guidance_lines = []
        for line in lines:
            line_lower = line.lower()
            if any(word in line_lower for word in ['guidance', 'expect', 'outlook', 'forecast', 'target']):
                guidance_lines.append(line)
        
        if guidance_lines:
            for line in guidance_lines:
                talking_points.append(f"- {line}")
        else:
            talking_points.append("- We remain confident in our strategic direction and growth trajectory")
            talking_points.append("- Focused on executing our roadmap and delivering shareholder value")
        
        talking_points.append("\n### Closing\n")
        talking_points.append(f"Thank you for your continued support. We're excited about {company_name}'s future and look forward to your questions.")
    
    return "\n".join(talking_points)


def generate_talking_points(content, company_name, model_name="google/flan-t5-base", memo_style=False):
    """
    Generate management talking points for investor communications.
    
    Args:
        content: Source content (quarter summary, context, etc.)
        company_name: Name of the company
        model_name: HuggingFace model to use
        memo_style: If True, format as memo; if False, format as earnings call script
    
    Returns:
        Formatted talking points
    """
    generator = get_generator(model_name)
    
    if generator is None:
        # Use rule-based fallback
        return rule_based_talking_points(content, company_name, memo_style)
    
    # Use AI model
    if memo_style:
        prompt = f"""Create executive talking points for an investor memo about {company_name}.

Context:
{content[:2000]}

Generate professional talking points with:
1. Executive summary
2. Key strategic points
3. Investment considerations

Talking Points:"""
    else:
        prompt = f"""Create management talking points for {company_name}'s earnings call.

Content:
{content[:2000]}

Generate structured talking points with:
1. Opening remarks
2. Performance highlights
3. Strategic initiatives
4. Addressing challenges
5. Forward outlook
6. Closing remarks

Talking Points:"""
    
    try:
        result = generator(
            prompt,
            max_length=512,
            min_length=100,
            do_sample=True,
            temperature=0.7,
            top_p=0.9
        )
        talking_points = result[0]['generated_text']
        
        return talking_points
    
    except Exception as e:
        # Fallback to rule-based on any error
        return rule_based_talking_points(content, company_name, memo_style)