app/utils.py

from ultralytics import YOLO
from groq import Groq
from .config import MODEL_PATH, MODEL_IMAGE_SIZE, GROQ_API_KEY
import cv2
import json
import logging
import time

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Load model once
logger.info(f"Loading model from {MODEL_PATH}")
try:
    model = YOLO(MODEL_PATH)
except Exception as e:
    logger.error(f"Failed to load model: {e}")
    raise

def run_inference(image_path, output_path):
    """
    Runs YOLO inference on the image.
    Saves the annotated image to output_path.
    Returns a dictionary with detections and metadata.
    """
    start_time = time.time()
    
    # Run inference
    results = model(image_path, imgsz=MODEL_IMAGE_SIZE, conf=0.4)
    
    end_time = time.time()
    processing_time_ms = (end_time - start_time) * 1000
    
    result = results[0]
    
    # Save annotated image
    annotated_frame = result.plot()
    cv2.imwrite(str(output_path), annotated_frame)
    
    # Severity Mapping Logic
    SEVERITY_MAP = {
        "scratch": "Low",
        "dent": "Medium",
        "crack": "High",
        "lamp_broken": "High",
        "tire_flat": "High",
        "glass_shatter": "Critical"
    }

    # Extract detections for LLM and UI
    detections = []
    for box in result.boxes:
        cls_id = int(box.cls[0])
        class_name = result.names[cls_id]
        confidence = float(box.conf[0])
        
        # Simple location logic based on box center
        x_center = box.xywh[0][0]
        img_width = result.orig_shape[1]
        if x_center < img_width / 3:
            location = "Left Side"
        elif x_center > 2 * img_width / 3:
            location = "Right Side"
        else:
            location = "Center"

        # Determine Severity
        severity = SEVERITY_MAP.get(class_name, "Medium")
        
        # Boost severity if confidence is very high for critical items
        if severity == "Critical" and confidence > 0.8:
            severity = "CRITICAL (Safety Risk)"

        detections.append({
            "class": class_name,
            "confidence": confidence,
            "location": location,
            "severity": severity 
        })
        
    return {
        "detections": detections,
        "processing_time": round(processing_time_ms, 2),
        "count": len(detections),
        "model_info": f"YOLO11s ({MODEL_IMAGE_SIZE})"
    }

def generate_insurance_report(detection_data, make, model_name, year):
    """
    Generates a structured JSON report using Groq API based on detections.
    """
    if not GROQ_API_KEY:
        return {"error": "Groq API Key not configured"}
        
    client = Groq(api_key=GROQ_API_KEY)
    
    vehicle_info = f"{year} {make} {model_name}"
    detections = detection_data.get("detections", [])
    det_str = json.dumps(detections, indent=2)
    
    # Fallback if no detections
    if not detections:
        return {
            "summary": f"No specific damage patterns were detected by the automated system on the {vehicle_info}.",
            "severity_level": "Low",
            "severity_reasoning": "No visible dents, scratches, or tears identified by AI.",
            "repair_urgency": "None",
            "urgency_details": "Vehicle appears to be in good condition based on visual inspection.",
            "cost_range": "$0",
            "cost_details": "No repairs needed.",
            "recommended_action": "Routine maintenance.",
            "immediate_actions": ["Verify visual inspection manually.", "Wash vehicle to ensure no hidden marks."],
            "secondary_actions": ["Keep regular service schedule."]
        }

    prompt = f"""
    Act as a Senior Automotive Claims Adjuster and Technical Appraiser. 
    Analyze the following vehicle damage detection data provided by an automated computer vision system.
    
    Vehicle: {vehicle_info}
    Detected Damages: {det_str}
    
    Generate a formal, industry-standard damage assessment report in valid JSON format ONLY. 
    Use professional terminology (e.g., 'refinish', 'R&I', 'structural integrity') and maintain an objective tone.
    
    The JSON must have these exact keys:
    {{
        "summary": "A formal executive summary of the visual inspection findings.",
        "severity_level": "Low, Medium, High, or Total Loss",
        "severity_reasoning": "Technical justification for the severity classification.",
        "repair_urgency": "Routine, Urgent, or Safety Critical",
        "urgency_details": "Explanation focusing on safety and drivability.",
        "cost_range": "Estimated repair cost range (e.g., '$800 - $1,500 USD').",
        "cost_details": "Breakdown of cost drivers (e.g., 'Includes body labor, paint supplies, and parts').",
        "estimated_labor_hours": "Estimated labor time (e.g., '12-16 hours').",
        "likely_parts_needed": ["List of parts likely requiring repair or replacement"],
        "recommended_action": "Primary technical recommendation (e.g., 'Tear-down and blueprinting').",
        "immediate_actions": ["Action 1", "Action 2"],
        "secondary_actions": ["Action 1", "Action 2"]
    }}
    """
        
    try:
        chat_completion = client.chat.completions.create(
            messages=[
                {
                    "role": "user",
                    "content": prompt,
                }
            ],
            model="llama-3.3-70b-versatile",
            response_format={"type": "json_object"} # Enforce JSON mode
        )
        return json.loads(chat_completion.choices[0].message.content)
    except Exception as e:
        logger.error(f"Groq API error: {e}")
        return {
            "summary": "Error generating detailed report.",
            "severity_level": "Unknown",
            "severity_reasoning": str(e),
            "repair_urgency": "Unknown",
            "urgency_details": "Contact support.",
            "cost_range": "Unknown",
            "cost_details": "Unknown",
            "estimated_labor_hours": "Unknown",
            "likely_parts_needed": [],
            "recommended_action": "Manual Inspection",
            "immediate_actions": ["Contact support"],
            "secondary_actions": []
        }