src/compiler.py

"""
Response Compiler - Stage 3
Compiles MCP results with focus on ALERTING INFORMATION ONLY
"""

from typing import Dict, Any
from openai import OpenAI
import httpx
import os


class ResponseCompiler:
    """
    Compiles MCP server results into alert-focused responses.
    
    KEY PRINCIPLE: All MCPs are queried, but the compiler extracts ONLY
    the alerting, concerning, or actionable information. Normal/good status
    is minimized or omitted entirely.
    """
    
    def __init__(self):
        """Initialize compiler with OpenAI client"""
        api_key = os.getenv("OPENAI_API_KEY")
        if not api_key:
            raise ValueError("OPENAI_API_KEY environment variable not set")
        
        # Create custom httpx client without proxies parameter (compatibility fix)
        http_client = httpx.Client(
            timeout=httpx.Timeout(60.0, connect=10.0),
            limits=httpx.Limits(max_keepalive_connections=10, max_connections=20)
        )
        
        self.client = OpenAI(api_key=api_key, http_client=http_client)
    
    def compile_alert_summary(
        self, 
        mcp_results: Dict[str, Any], 
        location: Dict[str, float],
        location_name: str = ""
    ) -> str:
        """
        Compile MCP results into alert summary focusing ONLY on concerning information.
        
        This is where the intelligence lives - not in routing. All MCP servers are
        queried, but we extract only what farmers need to act on.
        
        Args:
            mcp_results: Results from all MCP servers
            location: Dict with 'latitude' and 'longitude' keys
            location_name: Optional human-readable location name
            
        Returns:
            Alert summary highlighting only actionable concerns
        """
        
        # Build comprehensive context from all MCP data
        context_parts = []
        
        for server_name, result in mcp_results.items():
            if result["status"] == "success" and result["data"]:
                context_parts.append(f"=== {server_name.upper()} DATA ===\n{result['data']}")
        
        if not context_parts:
            return "Unable to generate alert summary - no data available from MCP servers."
        
        full_context = "\n\n".join(context_parts)
        location_str = f"{location_name} ({location['latitude']:.4f}°N, {location['longitude']:.4f}°E)"
        
        # THE KEY PROMPT: Extract only alerting information
        prompt = f"""You are an agricultural alert analyst. Your task is to analyze comprehensive agricultural data and extract ONLY the alerting, concerning, or time-sensitive information.

LOCATION: {location_str}

COMPREHENSIVE DATA FROM ALL MONITORING SYSTEMS:
{full_context}

YOUR TASK:
Generate a concise ALERT SUMMARY that includes ONLY:

1. **CRITICAL ALERTS** - Immediate threats requiring urgent action:
   - Extreme weather conditions (heat waves, storms, frost)
   - Active pest/disease outbreaks
   - Severe water scarcity or excess
   - Soil contamination or extreme deficiencies
   
2. **IMPORTANT WARNINGS** - Developing issues requiring attention:
   - Concerning trends (declining water table, degrading soil)
   - Moderate pest pressure building up
   - Suboptimal weather patterns affecting crops
   - Nutrient imbalances needing correction

3. **ACTIONABLE RECOMMENDATIONS** - What farmers should do:
   - Specific actions with timing
   - Preventive measures
   - Mitigation strategies

CRITICAL RULES:
- OMIT all normal/good status information unless it provides important context
- If weather is normal → DON'T mention it or say "Weather: Normal" briefly
- If soil is healthy → SKIP or say "Soil: No concerns" briefly  
- If no pest activity → SKIP or say "Pests: No threats detected" briefly
- FOCUS on deviations from normal, risks, and time-sensitive items
- Use specific numbers/dates only when they convey urgency
- Maximum 400 words total
- If everything is fine, say so clearly upfront then provide brief context

Structure:
1. Status Line: "CRITICAL ALERTS DETECTED" or "NO CRITICAL ALERTS - FAVORABLE CONDITIONS"
2. Critical Alerts section (if any)
3. Important Warnings section (if any)
4. Recommended Actions (always include if alerts/warnings exist)
5. Add raw API output in JSON format at end for reference.

Be direct. Skip pleasantries. Farmers need to know what matters."""

        try:
            response = self.client.chat.completions.create(
                model="gpt-4o",
                messages=[
                    {
                        "role": "system", 
                        "content": "You are an expert agricultural alert analyst. Extract ONLY concerning, alerting, or actionable information. Omit normal status unless contextually necessary."
                    },
                    {"role": "user", "content": prompt}
                ],
                temperature=0.2,
                max_tokens=1000
            )
            
            return response.choices[0].message.content.strip()
            
        except Exception as e:
            print(f"⚠️ Compilation error: {e}")
            return self._create_fallback_summary(mcp_results, location_str)
    
    def compile_response(
        self, 
        query: str, 
        mcp_results: Dict[str, Any], 
        location: Dict[str, float]
    ) -> str:
        """
        Compile MCP results into a response for a specific query.
        
        Args:
            query: User's original query
            mcp_results: Results from MCP servers
            location: Dict with 'latitude' and 'longitude' keys
            
        Returns:
            Compiled response text focusing on query-relevant information
        """
        
        # Format MCP results for context
        context_parts = []
        for server_name, result in mcp_results.items():
            if result["status"] == "success" and result["data"]:
                context_parts.append(f"{server_name.upper()}: {result['data']}")
        
        context = "\n\n".join(context_parts)
        
        prompt = f"""Answer this farmer's question using the provided data, focusing on actionable insights.

QUESTION: {query}
LOCATION: {location['latitude']:.4f}°N, {location['longitude']:.4f}°E

AVAILABLE DATA:
{context}

Provide a focused answer that:
1. Directly addresses the question
2. Highlights any concerning information relevant to the query
3. Gives specific recommendations
4. Keeps explanations brief and practical
5. Omits irrelevant normal/good status information
6. Add Raw API Output from all MCP Servers at the end for reference.

Be conversational but professional. Skip unnecessary background unless it aids understanding."""

        try:
            response = self.client.chat.completions.create(
                model="gpt-4o",
                messages=[
                    {"role": "system", "content": "You are a knowledgeable agricultural advisor providing practical guidance to farmers."},
                    {"role": "user", "content": prompt}
                ],
                temperature=0.5,
                max_tokens=800
            )
            
            return response.choices[0].message.content.strip()
            
        except Exception as e:
            print(f"⚠️ Compilation error: {e}")
            return f"Error compiling response: {str(e)}"
    
    def _create_fallback_summary(self, mcp_results: Dict[str, Any], location_str: str) -> str:
        """Create basic fallback summary if LLM compilation fails"""
        
        summary_parts = [f"Alert Summary for {location_str}\n\n"]
        
        for server_name, result in mcp_results.items():
            if result["status"] == "success" and result["data"]:
                summary_parts.append(f"{server_name.upper()}:")
                summary_parts.append(str(result["data"])[:300] + "...\n")
        
        return "\n".join(summary_parts)