"""
Autonomous Response Enhancer - 100% LLM-Driven
===============================================

NO hardcoded enhancements!
Everything is generated by LLM dynamically.

Features:
- Autonomous insight generation
- Dynamic follow-up suggestions
- Context-aware tone adjustment
- Data-specific formatting
"""

import json
import logging
from typing import Dict, List, Optional

from core.llm import chat

logger = logging.getLogger(__name__)


def extract_data_summary_from_response(response: str, currency_symbol: str = "$") -> Dict:
    """
    Extract data summary autonomously using LLM.
    """
    try:
        prompt = f"""Analyze this data analysis response and extract key metrics.

RESPONSE: "{response[:600]}"

Return JSON with extracted data:
{{
    "key_numbers": ["list of important numbers found"],
    "percentages": ["any percentages mentioned"],
    "entities": ["entities/names mentioned"],
    "main_finding": "one sentence summary of main finding"
}}

JSON:"""

        result = chat(prompt, temperature=0.1, max_tokens=150)
        
        # Parse JSON
        result = result.strip()
        if '```' in result:
            result = result.split('```')[1]
            if result.startswith('json'):
                result = result[4:]
        
        start = result.find('{')
        end = result.rfind('}') + 1
        if start >= 0 and end > start:
            result = result[start:end]
        
        return json.loads(result)
    except:
        return {"key_numbers": [], "percentages": [], "entities": [], "main_finding": ""}


def enhance_with_insight(
    response: str,
    query: str,
    data_context: str = ""
) -> str:
    """
    Add insight fully autonomously - NO hardcoded patterns!
    """
    if len(response) < 100 or "💡" in response:
        return response
    
    try:
        prompt = f"""Based on this data analysis, generate ONE specific insight.

QUERY: {query}
RESPONSE: {response[:500]}

The insight should be:
- Specific to THIS data (not generic advice)
- Start with 💡
- Be 1-2 sentences max
- Provide actionable or surprising information

Generate the insight (just the insight text, starting with 💡):"""

        insight = chat(prompt, temperature=0.7, max_tokens=80)
        insight = insight.strip()
        
        if insight and len(insight) > 10:
            return response + f"\n\n{insight}"
    except Exception as e:
        logger.debug(f"Insight generation error: {e}")
    
    return response


def enhance_with_suggestions(
    response: str,
    query: str,
    columns: List[str] = None
) -> str:
    """
    Add follow-up suggestions fully autonomously - NO hardcoding!
    """
    if len(response) < 100 or "You might also" in response:
        return response
    
    try:
        prompt = f"""Based on this analysis, suggest 2 natural follow-up questions.

QUERY: {query}
RESPONSE: {response[:400]}
DATA COLUMNS: {columns or "Unknown"}

Generate exactly 2 follow-up questions that would be logical next steps.
Format as:
1. [first question]
2. [second question]

Questions:"""

        result = chat(prompt, temperature=0.7, max_tokens=100)
        
        # Parse questions
        lines = result.strip().split('\n')
        questions = []
        for line in lines:
            line = line.strip()
            if line and (line[0].isdigit() or line.startswith('-') or line.startswith('•')):
                # Remove numbering
                q = line.lstrip('0123456789.-•) ').strip()
                if q and len(q) > 5:
                    questions.append(q)
        
        if questions:
            suggestion_text = "\n\n---\n**You might also ask:**\n"
            for q in questions[:2]:
                suggestion_text += f"• {q}\n"
            return response + suggestion_text
    except Exception as e:
        logger.debug(f"Suggestion generation error: {e}")
    
    return response


def enhance_tone(response: str, query: str) -> str:
    """
    Enhance tone autonomously - NO hardcoded replacements!
    """
    # Only enhance longer responses
    if len(response) < 200:
        return response
    
    # Check if tone seems robotic
    robotic_indicators = ['Based on the data provided', 'According to the information', 
                         'It can be observed', 'The analysis indicates']
    
    needs_enhancement = any(ind in response for ind in robotic_indicators)
    
    if not needs_enhancement:
        return response
    
    try:
        prompt = f"""Rewrite this response to be more natural and conversational, like ChatGPT.
Keep all the data and facts exactly the same.
Just make the tone warmer and more engaging.

ORIGINAL RESPONSE:
{response[:800]}

REWRITTEN (keep same facts, warmer tone):"""

        enhanced = chat(prompt, temperature=0.5, max_tokens=800)
        
        if enhanced and len(enhanced) > len(response) * 0.5:
            return enhanced.strip()
    except:
        pass
    
    return response


def enhance_full_response(
    query: str,
    response: str,
    query_type: str = "general",
    data_summary: Dict = None,
    entities: List[str] = None,
    add_insight: bool = True,
    add_suggestions: bool = True,
    enhance_tone_flag: bool = True,
    currency_symbol: str = "$",
    domain: str = "general",
    columns: List[str] = None
) -> str:
    """
    Fully autonomous response enhancement.
    NO hardcoded patterns - everything LLM-driven!
    """
    enhanced = response
    
    # Enhance tone first
    if enhance_tone_flag:
        enhanced = enhance_tone(enhanced, query)
    
    # Add autonomous insight
    if add_insight:
        enhanced = enhance_with_insight(enhanced, query)
    
    # Add autonomous suggestions
    if add_suggestions:
        enhanced = enhance_with_suggestions(enhanced, query, columns)
    
    return enhanced


def generate_autonomous_summary(
    data_context: str,
    columns: List[str],
    num_rows: int
) -> str:
    """
    Generate data summary fully autonomously.
    """
    try:
        prompt = f"""Generate a brief, helpful summary of this dataset.

COLUMNS: {columns}
ROWS: {num_rows}
SAMPLE DATA: {data_context[:500]}

Generate 2-3 sentences describing:
1. What kind of data this is
2. What analysis would be valuable

Summary:"""

        result = chat(prompt, temperature=0.5, max_tokens=150)
        return result.strip()
    except:
        return f"Dataset with {num_rows} rows and {len(columns)} columns."