app.py

from flask import Flask, render_template, request, Response
import requests
import datetime
from twilio.rest import Client  # For Twilio integration
from geopy.geocoders import Photon
from geopy.exc import GeocoderTimedOut, GeocoderServiceError
from transformers import pipeline
import warnings

# Suppress warnings
warnings.filterwarnings('ignore')

app = Flask(__name__)

# Initialize Photon geocoder (no API key required)
photon_geolocator = Photon(user_agent="MyWeatherApp", timeout=10)

# Initialize LLM for personalized recommendations
print("Loading LLM model for personalized recommendations...")
try:
    llm_generator = pipeline(
        "text-generation",
        model="distilgpt2",  # Lightweight model
        max_length=200,
        device=-1  # CPU
    )
    print("✅ LLM model loaded successfully!")
except Exception as e:
    print(f"⚠️ LLM model loading failed: {e}")
    llm_generator = None

def parse_iso_datetime(timestr):
    """
    Parse an ISO8601 datetime string (removing any trailing 'Z').
    """
    if timestr.endswith("Z"):
        timestr = timestr[:-1]
    return datetime.datetime.fromisoformat(timestr)

def find_closest_hour_index(hour_times, current_time_str):
    """
    Find the index in hour_times that is closest to the current_time_str.
    """
    if not hour_times:
        return None
    dt_current = parse_iso_datetime(current_time_str)
    min_diff = None
    best_index = None
    for i, ht in enumerate(hour_times):
        dt_ht = parse_iso_datetime(ht)
        diff = abs((dt_ht - dt_current).total_seconds())
        if min_diff is None or diff < min_diff:
            min_diff = diff
            best_index = i
    return best_index

def get_weather_icon(code):
    """Map the Open-Meteo weathercode to an emoji icon."""
    if code == 0:
        return "☀️"  # Clear sky
    elif code in [1, 2, 3]:
        return "⛅"
    elif code in [45, 48]:
        return "🌫️"
    elif code in [51, 53, 55]:
        return "🌦️"
    elif code in [56, 57]:
        return "🌧️"
    elif code in [61, 63, 65]:
        return "🌧️"
    elif code in [66, 67]:
        return "🌧️"
    elif code in [71, 73, 75, 77]:
        return "❄️"
    elif code in [80, 81, 82]:
        return "🌦️"
    elif code in [85, 86]:
        return "❄️"
    elif code in [95, 96, 99]:
        return "⛈️"
    else:
        return "❓"

def get_weather_description(code):
    """Short textual description for the weathercode."""
    descriptions = {
        0: "Clear sky",
        1: "Mainly clear",
        2: "Partly cloudy",
        3: "Overcast",
        45: "Fog",
        48: "Depositing rime fog",
        51: "Light drizzle",
        53: "Moderate drizzle",
        55: "Dense drizzle",
        56: "Freezing drizzle",
        57: "Freezing drizzle",
        61: "Slight rain",
        63: "Moderate rain",
        65: "Heavy rain",
        66: "Freezing rain",
        67: "Freezing rain",
        71: "Slight snow fall",
        73: "Moderate snow fall",
        75: "Heavy snow fall",
        77: "Snow grains",
        80: "Slight rain showers",
        81: "Moderate rain showers",
        82: "Violent rain showers",
        85: "Slight snow showers",
        86: "Heavy snow showers",
        95: "Thunderstorm",
        96: "Thunderstorm w/ slight hail",
        99: "Thunderstorm w/ heavy hail"
    }
    return descriptions.get(code, "Unknown")

def reverse_geocode(lat, lon):
    """
    Use Photon (via geopy) to convert latitude and longitude into a human-readable address.
    If the geocoding fails, returns a fallback string with the coordinates.
    """
    try:
        location = photon_geolocator.reverse((lat, lon), exactly_one=True)
        if location:
            return location.address
    except (GeocoderTimedOut, GeocoderServiceError) as e:
        print("Photon reverse geocode error:", e)
    return f"Lat: {lat}, Lon: {lon}"

# -----------------------------
# LLM-Powered Personalized Recommendations
# -----------------------------
def generate_personalized_recommendations(weather_summary, location_address, critical_days, warning_days):
    """
    Generate personalized agricultural recommendations using LLM based on weather and location.
    """
    if llm_generator is None:
        return None
    
    try:
        # Extract region info from location
        region_parts = location_address.split(',')
        region = region_parts[-1].strip() if len(region_parts) > 0 else "your region"
        
        # Determine weather condition
        weather_condition = "cloudy conditions"
        if critical_days:
            if len(critical_days) > 3:
                weather_condition = "severe weather alerts"
            else:
                weather_condition = "critical weather conditions"
        elif warning_days:
            weather_condition = "warning-level weather"
        
        # Create a better structured prompt
        prompt = f"""Agricultural advice for farmers in {region}:

Weather: {weather_condition} expected
Days affected: {len(critical_days) + len(warning_days)} days

Farming recommendations:
1. Crop care:"""

        # Generate recommendations
        response = llm_generator(
            prompt,
            max_new_tokens=80,
            num_return_sequences=1,
            temperature=0.8,
            do_sample=True,
            pad_token_id=50256,
            repetition_penalty=1.5  # Reduce repetition
        )
        
        generated_text = response[0]['generated_text']
        # Extract only the generated part
        recommendations = generated_text[len(prompt):].strip()
        
        # Clean up the output
        lines = recommendations.split('\n')
        clean_lines = []
        seen = set()
        
        for line in lines[:5]:  # Max 5 lines
            line = line.strip()
            # Skip empty, repetitive, or nonsensical lines
            if line and len(line) > 10 and line not in seen:
                # Check for repetition patterns
                if not any(line.count(word) > 2 for word in line.split()):
                    clean_lines.append(line)
                    seen.add(line)
        
        if clean_lines:
            recommendations = ' '.join(clean_lines)
            # Limit length
            if len(recommendations) > 180:
                recommendations = recommendations[:177] + "..."
            return recommendations
        else:
            # Fallback to rule-based if LLM output is poor
            return generate_rule_based_recommendations(weather_condition, region, critical_days, warning_days)
        
    except Exception as e:
        print(f"LLM generation error: {e}")
        return generate_rule_based_recommendations("cloudy conditions", "your region", critical_days, warning_days)

def generate_rule_based_recommendations(weather_condition, region, critical_days, warning_days):
    """
    Fallback rule-based recommendations when LLM fails or produces poor output.
    """
    if critical_days and len(critical_days) > 0:
        return f"For {region}: Secure crops and equipment. Postpone spraying. Monitor drainage systems. Harvest ready crops before severe weather."
    elif warning_days and len(warning_days) > 0:
        if "cloudy" in weather_condition.lower() or "overcast" in weather_condition.lower():
            return f"For {region}: Reduce irrigation due to lower evaporation. Monitor for fungal diseases. Apply preventive fungicides if needed."
        else:
            return f"For {region}: Adjust irrigation schedule. Monitor soil moisture. Delay non-essential field operations."
    else:
        return f"For {region}: Continue normal farming operations. Monitor weather updates regularly."

# -----------------------------
# Twilio WhatsApp Integration
# -----------------------------
def check_and_collect_alerts(forecast_list):
    """
    Check the forecast for hazardous weather conditions and collect detailed alert messages.
    """
    alerts = []
    critical_days = []
    warning_days = []
    
    for day in forecast_list:
        day_alerts = []
        severity = "INFO"
        
        # Temperature Analysis
        if day.get("tmax") and day.get("tmin"):
            tmax = day["tmax"]
            tmin = day["tmin"]
            
            if tmax > 40:
                day_alerts.append(f"🌡️ Extreme Heat: {tmax}°C - High risk of crop stress and water loss")
                severity = "CRITICAL"
            elif tmax > 35:
                day_alerts.append(f"🌡️ High Temperature: {tmax}°C - Increase irrigation frequency")
                severity = "WARNING"
            
            if tmin < 5:
                day_alerts.append(f"❄️ Frost Risk: {tmin}°C - Protect sensitive crops")
                severity = "CRITICAL"
            elif tmin < 10:
                day_alerts.append(f"🌡️ Cold Night: {tmin}°C - Monitor young plants")
                severity = "WARNING"
            
            # Temperature swing
            temp_diff = tmax - tmin
            if temp_diff > 20:
                day_alerts.append(f"📊 Large temperature swing: {temp_diff}°C - May stress plants")
        
        # Weather Condition Analysis
        desc = day.get("desc", "").lower()
        if "thunderstorm" in desc or "heavy" in desc:
            day_alerts.append(f"⛈️ Severe Weather: {day['desc']} - Secure equipment, delay spraying")
            severity = "CRITICAL"
        elif "rain" in desc or "drizzle" in desc:
            day_alerts.append(f"🌧️ Rainfall Expected: {day['desc']} - Postpone irrigation, avoid field work")
            severity = "WARNING"
        elif "overcast" in desc or "cloudy" in desc:
            day_alerts.append(f"☁️ Cloudy Conditions: {day['desc']} - Reduced photosynthesis, monitor for diseases")
            severity = "INFO"
        
        # Compile day alert if any conditions met
        if day_alerts:
            day_header = f"\n*{day['day_name']} ({day['date_str']})*"
            if severity == "CRITICAL":
                day_header = f"🚨 {day_header} - CRITICAL"
                critical_days.append(day['day_name'])
            elif severity == "WARNING":
                day_header = f"⚠️ {day_header} - WARNING"
                warning_days.append(day['day_name'])
            
            alert_text = day_header + "\n" + "\n".join(f"  • {alert}" for alert in day_alerts)
            
            # Add temperature range
            if day.get("tmax") and day.get("tmin"):
                alert_text += f"\n  📈 Temp Range: {day['tmin']}°C - {day['tmax']}°C"
            if day.get("morning_temp"):
                alert_text += f"\n  🌅 Morning: {day['morning_temp']}°C"
            if day.get("evening_temp"):
                alert_text += f"\n  🌆 Evening: {day['evening_temp']}°C"
            
            alerts.append(alert_text)
    
    return alerts, critical_days, warning_days

def send_whatsapp_message(message, location, location_address, critical_days, warning_days):
    """
    Send a WhatsApp message using Twilio API with enhanced agricultural insights.
    """
    google_maps_url = f"https://www.google.com/maps?q={location[0]},{location[1]}"
    
    # Build concise severity summary
    severity_summary = ""
    if critical_days:
        severity_summary += f"🚨 {len(critical_days)} CRITICAL: {', '.join(critical_days[:3])}\n"
    if warning_days:
        severity_summary += f"⚠️ {len(warning_days)} WARNING: {', '.join(warning_days[:3])}\n"
    
    # Generate personalized LLM recommendations (shortened)
    weather_summary = f"{len(critical_days)} critical, {len(warning_days)} warning days"
    llm_recommendations = generate_personalized_recommendations(
        weather_summary, 
        location_address, 
        critical_days, 
        warning_days
    )
    
    # Shorten LLM recommendations if too long
    if llm_recommendations and len(llm_recommendations) > 200:
        llm_recommendations = llm_recommendations[:197] + "..."
    
    # Build concise message
    message_content = (
        f"🌾 *WEATHER ALERT* 🌾\n\n"
        f"{severity_summary}\n"
    )
    
    # Add condensed forecast (max 3 days)
    alert_lines = message.strip().split('\n\n')[:3]
    for alert in alert_lines:
        # Shorten each alert
        lines = alert.split('\n')
        if lines:
            message_content += f"{lines[0]}\n"  # Just the header
            if len(lines) > 1:
                message_content += f"{lines[1][:80]}\n"  # First detail only
    
    message_content += "\n"
    
    # Add AI recommendations if available
    if llm_recommendations:
        message_content += f"🤖 *AI ADVICE:*\n{llm_recommendations}\n\n"
    
    # Add critical actions only
    if critical_days:
        message_content += (
            f"🚨 *URGENT:*\n"
            f"• Secure equipment\n"
            f"• Harvest ready crops\n"
            f"• Protect livestock\n\n"
        )
    elif warning_days:
        message_content += (
            f"⚠️ *ACTIONS:*\n"
            f"• Adjust irrigation\n"
            f"• Monitor soil moisture\n"
            f"• Delay field work\n\n"
        )
    
    # Add location
    message_content += (
        f"📍 {location_address}\n"
        f"🗺️ {google_maps_url}\n\n"
        f"_Weather Forecast for Farmers_"
    )
    
    # Ensure under 1600 characters
    if len(message_content) > 1590:
        message_content = message_content[:1587] + "..."
    
    account_sid = 'ACe45f7038c5338a153d1126ca6d547c84'
    auth_token = '48b9eea898885ef395d48edc74924340'
    client = Client(account_sid, auth_token)
    
    try:
        msg = client.messages.create(
            from_='whatsapp:+14155238886',
            body=message_content,
            to='whatsapp:+919763059811'
        )
        print(f"✅ WhatsApp sent! SID: {msg.sid}, Length: {len(message_content)} chars")
    except Exception as e:
        print(f"❌ WhatsApp error: {e}")
        print(f"Message length was: {len(message_content)} characters")

@app.route("/", methods=["GET", "POST"])
def index():
    # Default coordinates
    default_lat = 18.5196
    default_lon = 73.8553

    if request.method == "POST":
        try:
            lat = float(request.form.get("lat", default_lat))
            lon = float(request.form.get("lon", default_lon))
        except ValueError:
            lat, lon = default_lat, default_lon
    else:
        lat = float(request.args.get("lat", default_lat))
        lon = float(request.args.get("lon", default_lon))

    location_address = reverse_geocode(lat, lon)

    # Call Open-Meteo API for forecast data
    url = "https://api.open-meteo.com/v1/forecast"
    params = {
        "latitude": lat,
        "longitude": lon,
        "hourly": (
            "temperature_2m,relative_humidity_2m,precipitation,"
            "cloudcover,windspeed_10m,pressure_msl,soil_moisture_3_to_9cm,uv_index"
        ),
        "daily": (
            "weathercode,temperature_2m_max,temperature_2m_min,"
            "sunrise,sunset,uv_index_max"
        ),
        "current_weather": True,
        "forecast_days": 10,
        "timezone": "auto"
    }
    resp = requests.get(url, params=params)
    data = resp.json()

    timezone = data.get("timezone", "Local")
    current_weather = data.get("current_weather", {})
    current_temp = current_weather.get("temperature")
    current_time = current_weather.get("time")
    current_code = current_weather.get("weathercode")
    current_icon = get_weather_icon(current_code)
    current_desc = get_weather_description(current_code)
    current_wind_speed = current_weather.get("windspeed", 0.0)
    current_wind_dir = current_weather.get("winddirection", 0)

    if current_time:
        dt_current = parse_iso_datetime(current_time)
        current_time_formatted = dt_current.strftime("%A, %b %d, %Y %I:%M %p")
    else:
        current_time_formatted = ""

    hourly_data = data.get("hourly", {})
    hour_times = hourly_data.get("time", [])
    hour_temp = hourly_data.get("temperature_2m", [])
    hour_humidity = hourly_data.get("relative_humidity_2m", [])
    hour_precip = hourly_data.get("precipitation", [])
    hour_clouds = hourly_data.get("cloudcover", [])
    hour_wind = hourly_data.get("windspeed_10m", [])
    hour_pressure = hourly_data.get("pressure_msl", [])
    hour_soil = hourly_data.get("soil_moisture_3_to_9cm", [])
    hour_uv = hourly_data.get("uv_index", [])

    current_index = None
    if current_time:
        current_index = find_closest_hour_index(hour_times, current_time)

    feels_like = current_temp
    if current_index is not None and current_index < len(hour_humidity):
        h = hour_humidity[current_index]
        feels_like = round(current_temp - 0.2 * (100 - h) / 10, 1)

    today_highlights = {}
    if current_index is not None:
        today_highlights["humidity"] = hour_humidity[current_index] if current_index < len(hour_humidity) else None
        today_highlights["precipitation"] = hour_precip[current_index] if current_index < len(hour_precip) else None
        today_highlights["clouds"] = hour_clouds[current_index] if current_index < len(hour_clouds) else None
        today_highlights["windspeed"] = hour_wind[current_index] if current_index < len(hour_wind) else None
        today_highlights["pressure"] = hour_pressure[current_index] if current_index < len(hour_pressure) else None
        today_highlights["soil_moisture"] = hour_soil[current_index] if current_index < len(hour_soil) else None
        today_highlights["uv_index"] = hour_uv[current_index] if current_index < len(hour_uv) else None
    else:
        for k in ["humidity", "precipitation", "cloudcover", "windspeed", "pressure", "soil_moisture", "uv_index"]:
            today_highlights[k] = None

    daily_data = data.get("daily", {})
    daily_sunrise = daily_data.get("sunrise", [])
    daily_sunset = daily_data.get("sunset", [])
    if len(daily_sunrise) > 0:
        today_highlights["sunrise"] = daily_sunrise[0][11:16]
    else:
        today_highlights["sunrise"] = None
    if len(daily_sunset) > 0:
        today_highlights["sunset"] = daily_sunset[0][11:16]
    else:
        today_highlights["sunset"] = None

    daily_times = daily_data.get("time", [])
    daily_codes = daily_data.get("weathercode", [])
    daily_tmax = daily_data.get("temperature_2m_max", [])
    daily_tmin = daily_data.get("temperature_2m_min", [])
    forecast_list = []

    def get_hour_temp(date_str, hour_str):
        target = date_str + "T" + hour_str + ":00"
        best_idx = None
        best_diff = None
        dt_target = parse_iso_datetime(target)
        for i, ht in enumerate(hour_times):
            dt_ht = parse_iso_datetime(ht)
            diff = abs((dt_ht - dt_target).total_seconds())
            if best_diff is None or diff < best_diff:
                best_diff = diff
                best_idx = i
        if best_idx is not None and best_idx < len(hour_temp):
            return hour_temp[best_idx]
        return None

    for i in range(len(daily_times)):
        date_str = daily_times[i]
        dt_obj = parse_iso_datetime(date_str)
        day_name = dt_obj.strftime("%A")
        short_date = dt_obj.strftime("%b %d")

        code = daily_codes[i] if i < len(daily_codes) else None
        icon = get_weather_icon(code)
        desc = get_weather_description(code)

        tmax = daily_tmax[i] if i < len(daily_tmax) else None
        tmin = daily_tmin[i] if i < len(daily_tmin) else None
        avg_temp = round((tmax + tmin) / 2, 1) if tmax is not None and tmin is not None else None

        morning_temp = get_hour_temp(date_str, "09")
        evening_temp = get_hour_temp(date_str, "21")

        sr = daily_sunrise[i][11:16] if i < len(daily_sunrise) else None
        ss = daily_sunset[i][11:16] if i < len(daily_sunset) else None

        forecast_list.append({
            "day_name": day_name,
            "date_str": short_date,
            "icon": icon,
            "desc": desc,
            "avg_temp": avg_temp,
            "morning_temp": morning_temp,
            "evening_temp": evening_temp,
            "sunrise": sr,
            "sunset": ss,
            "tmax": tmax,
            "tmin": tmin
        })


    alerts, critical_days, warning_days = check_and_collect_alerts(forecast_list)
    
    # Generate AI recommendations for frontend display
    ai_recommendations = None
    if alerts or critical_days or warning_days:
        weather_summary = f"{len(critical_days)} critical days, {len(warning_days)} warning days"
        ai_recommendations = generate_personalized_recommendations(
            weather_summary, 
            location_address, 
            critical_days, 
            warning_days
        )
    
    # Send WhatsApp alerts if needed
    if alerts:
        alert_message = "\n".join(alerts)
        send_whatsapp_message(alert_message, (lat, lon), location_address, critical_days, warning_days)
        alerts_sent = True
    else:
        alerts_sent = False

    return render_template(
        "index.html",
        lat=lat,
        lon=lon,
        location_address=location_address,
        current_temp=current_temp,
        current_icon=current_icon,
        current_desc=current_desc,
        current_time=current_time_formatted,
        current_wind_speed=current_wind_speed,
        current_wind_dir=current_wind_dir,
        feels_like=feels_like,
        today_highlights=today_highlights,
        forecast_list=forecast_list,
        timezone=timezone,
        alerts_sent=alerts_sent,
        ai_recommendations=ai_recommendations,
        critical_days=critical_days,
        warning_days=warning_days
    )

if __name__ == "__main__":
    app.run(debug=True,port=5001)