app.py

import os
from flask import Flask, render_template, request, jsonify, send_file
import requests
import pandas as pd
from datetime import datetime
import plotly.express as px
import plotly.io as pio
import numpy as np
import dotenv
import json
import gtts
import uuid
from pathlib import Path

dotenv.load_dotenv()

app = Flask(__name__)

# Create audio directory if it doesn't exist
AUDIO_DIR = Path("static/audio")
AUDIO_DIR.mkdir(parents=True, exist_ok=True)

def fetch_market_data(state=None, district=None, market=None, commodity=None):
    """Fetch data from the agricultural market API.
    If the API fails or returns empty data, fallback to the CSV file.
    Filters (state, district, market, commodity) are applied manually on CSV data.
    """
    api_key = os.getenv('api_key')
    base_url = "1https://api.data.gov.in/resource/9ef84268-d588-465a-a308-a864a43d007"
    
    params = {
        "api-key": api_key,
        "format": "json",
        "limit": 1000,
    }
    
    if state:
        params["filters[state]"] = state
    if district:
        params["filters[district]"] = district
    if market:
        params["filters[market]"] = market
    if commodity:
        params["filters[commodity]"] = commodity

    try:
        response = requests.get(base_url, params=params)
        if response.status_code == 200:
            data = response.json()
            records = data.get("records", [])
            df = pd.DataFrame(records)
        else:
            print(f"API Error: {response.status_code}")
            raise Exception(f"API Error: {response.status_code}")
    except Exception as e:
        print(f"Error fetching data from API: {str(e)}. Falling back to CSV file.")
        df = pd.read_csv("final_price_data.csv")
        if 'min_price' not in df.columns:
            rename_mapping = {
                'State': 'state',
                'District': 'district',
                'Market': 'market',
                'Commodity': 'commodity',
                'Variety': 'variety',
                'Grade': 'grade',
                'Arrival_Date': 'arrival_date',
                'Min_x0020_Price': 'min_price',
                'Max_x0020_Price': 'max_price',
                'Modal_x0020_Price': 'modal_price'
            }
            df.rename(columns=rename_mapping, inplace=True)
    
    if df.empty:
        print("API returned empty data. Falling back to CSV file.")
        df = pd.read_csv("final_price_data.csv")
        if 'min_price' not in df.columns:
            rename_mapping = {
                'State': 'state',
                'District': 'district',
                'Market': 'market',
                'Commodity': 'commodity',
                'Variety': 'variety',
                'Grade': 'grade',
                'Arrival_Date': 'arrival_date',
                'Min_x0020_Price': 'min_price',
                'Max_x0020_Price': 'max_price',
                'Modal_x0020_Price': 'modal_price'
            }
            df.rename(columns=rename_mapping, inplace=True)
    
    if state:
        df = df[df['state'] == state]
    if district:
        df = df[df['district'] == district]
    if market:
        df = df[df['market'] == market]
    if commodity:
        df = df[df['commodity'] == commodity]
    
    return df

def get_ai_insights(market_data, state, district, market=None, commodity=None, language="English"):
    """Get enhanced insights from Gemini API with focus on profitable suggestions for farmers.
       Supports multiple languages through the prompt.
       Returns dynamic insights only. If something goes wrong, returns an empty string.
    """
    if not state or not district or market_data.empty:
        return ""
    
    try:
        # Filter data based on provided parameters
        district_data = market_data[market_data['district'] == district]
        if district_data.empty:
            return ""
        
        # Apply market filter if provided
        if market and not market_data[market_data['market'] == market].empty:
            market_specific = True
            district_data = district_data[district_data['market'] == market]
        else:
            market_specific = False
            
        # Apply commodity filter if provided
        if commodity and not market_data[market_data['commodity'] == commodity].empty:
            commodity_specific = True
            district_data = district_data[district_data['commodity'] == commodity]
        else:
            commodity_specific = False
        
        # Calculate price trends
        price_trends = district_data.groupby('commodity').agg({
            'modal_price': ['mean', 'min', 'max', 'std']
        }).round(2)
        
        # Using environment variable for Gemini API key
        api_key = os.getenv('GEMINI_API')
        if not api_key:
            print("Warning: Gemini API key not set")
            return ""
            
        price_trends['price_stability'] = (price_trends['modal_price']['std'] / 
                                         price_trends['modal_price']['mean']).round(2)
        
        district_data['arrival_date'] = pd.to_datetime(district_data['arrival_date'])
        district_data['month'] = district_data['arrival_date'].dt.month
        monthly_trends = district_data.groupby(['commodity', 'month'])['modal_price'].mean().round(2)
        
        market_competition = len(district_data['market'].unique())
        top_commodities = district_data.groupby('commodity')['modal_price'].mean().nlargest(5).index.tolist()
        
        # Get min and max prices for key commodities
        price_range_info = {}
        for commodity in top_commodities[:3]:
            comm_data = district_data[district_data['commodity'] == commodity]
            if not comm_data.empty:
                price_range_info[commodity] = {
                    'min': comm_data['modal_price'].min(),
                    'max': comm_data['modal_price'].max(),
                    'avg': comm_data['modal_price'].mean()
                }
        
        # Calculate market-specific metrics if market is selected
        market_details = ""
        if market_specific:
            market_details = f"""
            Market-specific information for {market}:
            - Number of commodities: {len(district_data['commodity'].unique())}
            - Most expensive commodity: {district_data.groupby('commodity')['modal_price'].mean().idxmax()}
            - Cheapest commodity: {district_data.groupby('commodity')['modal_price'].mean().idxmin()}
            """
        
        # Commodity-specific details if commodity is selected
        commodity_details = ""
        if commodity_specific:
            commodity_data = district_data[district_data['commodity'] == commodity]
            best_market = commodity_data.loc[commodity_data['modal_price'].idxmin()]['market']
            worst_market = commodity_data.loc[commodity_data['modal_price'].idxmax()]['market']
            
            commodity_details = f"""
            Commodity-specific information for {commodity}:
            - Best market to buy (lowest price): {best_market}
            - Highest priced market: {worst_market}
            - Price variance across markets: {commodity_data['modal_price'].std().round(2)}
            """
        
        # Improved prompt for better structured output with language support
        prompt = f"""
        Analyze the following agricultural market data for {district}, {state} and provide insights in {language} language.
        
        Market data:
        - Active markets: {market_competition}
        - Top crops: {', '.join(top_commodities[:5])}
        - Data from {len(price_trends.index)} crops and {len(monthly_trends)} monthly entries.
        
        Price information:
        {json.dumps(price_range_info, indent=2)}
        
        {market_details}
        {commodity_details}
        
        Analyze this data and provide insights about crop market trends and profitability. 
        Include specific numbers from the data about prices.
        
        Provide structured insights with clear sections. Use this exact format with bullet points:
        
        Crop Profitability Analysis:
        * [First insight about profitable crops with specific prices mentioned]
        * [Second insight]
        
        Market Price Analysis:
        * [First insight about markets with specific price ranges]
        * [Second insight]
        
        Recommendations for Farmers:
        * [Action item 1]
        * [Action item 2]
        """
        
        api_url = "https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-pro:generateContent"
        headers = {"Content-Type": "application/json"}
        
        payload = {
            "contents": [
                {
                    "parts": [
                        {"text": prompt}
                    ]
                }
            ],
            "generationConfig": {
                "temperature": 0.4,
                "maxOutputTokens": 1024
            }
        }
        
        response = requests.post(
            f"{api_url}?key={api_key}", 
            headers=headers, 
            json=payload, 
            timeout=20
        )
        
        if response.status_code == 200:
            response_data = response.json()
            if 'candidates' in response_data and len(response_data['candidates']) > 0:
                content = response_data['candidates'][0]['content']
                if 'parts' in content and len(content['parts']) > 0:
                    insights = content['parts'][0]['text']
                    return format_ai_insights(insights, language)
            print(f"API Response issue: {response.text[:100]}")
        else:
            print(f"Gemini API Error: {response.status_code} - {response.text[:100]}")
        
        return ""
    
    except Exception as e:
        print(f"Error generating insights: {str(e)}")
        return ""

def extract_text_from_insights(insights_html):
    """Extract pure text content from HTML insights for text-to-speech conversion."""
    # Simple HTML tag removal - for production, consider using BeautifulSoup for better parsing
    import re
    text = re.sub(r'<.*?>', ' ', insights_html)
    text = re.sub(r'\s+', ' ', text)  # Remove extra whitespace
    return text.strip()

def create_audio_from_text(text, language_code="en"):
    """Generate audio file from text using gTTS."""
    if not text:
        return None
    
    # Map UI language selection to gTTS language codes
    language_map = {
        "English": "en",
        "Hindi": "hi",
        "Tamil": "ta",
        "Telugu": "te",
        "Marathi": "mr",
        "Bengali": "bn",
        "Gujarati": "gu",
        "Kannada": "kn", 
        "Malayalam": "ml",
        "Punjabi": "pa"
    }
    
    tts_lang = language_map.get(language_code, "en")
    
    # Generate unique filename
    filename = f"{uuid.uuid4()}.mp3"
    filepath = AUDIO_DIR / filename
    
    try:
        tts = gtts.gTTS(text, lang=tts_lang, slow=False)
        tts.save(str(filepath))
        return f"/static/audio/{filename}"
    except Exception as e:
        print(f"Error creating audio: {str(e)}")
        return None

def create_audio_local_fallback(text, language_code="en"):
    """Local fallback for TTS when network is unavailable."""   
    try:
        # This requires pyttsx3 to be installed
        import pyttsx3
        engine = pyttsx3.init()
        
        # Generate unique filename
        filename = f"{uuid.uuid4()}.mp3"
        filepath = AUDIO_DIR / filename
        
        engine.save_to_file(text, str(filepath))
        engine.runAndWait()
        
        return f"/static/audio/{filename}"
    except Exception as e:
        print(f"Local TTS fallback failed: {str(e)}")
        return None

def format_ai_insights(insights_data, language="English"):
    """Format AI insights into structured HTML.
       Returns an empty string if no valid insights are provided.
    """
    if not insights_data or not insights_data.strip():
        return ""
    
    # Process the insights text - each bullet point becomes a formatted item
    formatted_content = ""
    
    # Split by bullet points
    bullet_points = insights_data.split('*')
    
    # Filter out empty items and process each bullet point
    bullet_points = [point.strip() for point in bullet_points if point.strip()]
    
    # Check if any section headers exist in the content
    sections = {}
    current_section = "Recommendations"
    
    for point in bullet_points:
        if ":" in point and len(point.split(":")[0]) < 30:  # Likely a section header
            current_section = point.split(":")[0].strip()
            # Start a new section
            if current_section not in sections:
                sections[current_section] = []
        else:
            # Add to current section
            if current_section not in sections:
                sections[current_section] = []
            sections[current_section].append(point)
    
    # Now build the HTML with proper sections
    for section, points in sections.items():
        formatted_content += f'<div class="insight-card"><h5>{section}</h5><ul class="insight-list">'
        for point in points:
            # Highlight prices with special styling
            if "₹" in point:
                # Replace price mentions with highlighted spans
                parts = point.split("₹")
                styled_point = parts[0]
                for i in range(1, len(parts)):
                    # Extract the price value
                    price_text = parts[i].split()[0]
                    # Add the highlighted price and the rest of the text
                    styled_point += f'<span class="price-highlight">₹{price_text}</span>' + parts[i][len(price_text):]
                formatted_content += f'<li>{styled_point}</li>'
            else:
                formatted_content += f'<li>{point}</li>'
        formatted_content += '</ul></div>'
    
    # Create the plain text version for audio generation
    plain_text = f"Market Insights for {language}.\n\n"
    for section, points in sections.items():
        plain_text += f"{section}:\n"
        for point in points:
            # Clean up for speech
            clean_point = point.replace("₹", " rupees ")
            plain_text += f"• {clean_point}\n"
        plain_text += "\n"
    
    # Generate audio file
    audio_path = create_audio_from_text(plain_text, language)
    if audio_path is None:
        audio_path = create_audio_local_fallback(plain_text)
    
    # Add a wrapper for the insights with audio player
    audio_player = ""
    if audio_path:
        audio_player = f"""
        <div class="audio-player-container">
            <h4>Listen to Insights</h4>
            <audio id="insightsAudio" controls>
                <source src="{audio_path}" type="audio/mpeg">
                Your browser does not support the audio element.
            </audio>
            <button class="btn btn-sm btn-custom mt-2" id="playAudioBtn">
                <i class="fa fa-play"></i> Play Audio
            </button>
        </div>
        """
    
    html = f"""
    <div class="insights-header">
        <h3>AI Market Insights</h3>
        {audio_player}
    </div>
    <div class="insight-section">
        {formatted_content}
    </div>
    """
    
    return html

def generate_plots(df):
    """Generate all plots in English"""
    if df.empty:
        return {}, "No data available"
    
    price_cols = ['min_price', 'max_price', 'modal_price']
    for col in price_cols:
        df[col] = pd.to_numeric(df[col], errors='coerce')
    
    colors = ["#4CAF50", "#8BC34A", "#CDDC39", "#FFC107", "#FF5722"]
    
    df_bar = df.groupby('commodity')['modal_price'].mean().reset_index()
    fig_bar = px.bar(df_bar, 
                     x='commodity', 
                     y='modal_price',
                     title="Average Price by Commodity",
                     color_discrete_sequence=colors)
    
    fig_line = None
    if 'commodity' in df.columns and len(df['commodity'].unique()) == 1:
        df['arrival_date'] = pd.to_datetime(df['arrival_date'])
        df_line = df.sort_values('arrival_date')
        fig_line = px.line(df_line, 
                          x='arrival_date', 
                          y='modal_price',
                          title="Price Trend",
                          color_discrete_sequence=colors)
    
    fig_box = px.box(df, 
                     x='commodity', 
                     y='modal_price',
                     title="Price Distribution",
                     color='commodity',
                     color_discrete_sequence=colors)
    
    plots = {
        'bar': pio.to_html(fig_bar, full_html=False),
        'box': pio.to_html(fig_box, full_html=False)
    }
    if fig_line:
        plots['line'] = pio.to_html(fig_line, full_html=False)
    
    return plots

@app.route('/')
def index():
    try:
        initial_data = fetch_market_data()
        states = sorted(initial_data['state'].dropna().unique()) if not initial_data.empty else []
    except Exception as e:
        print(f"Error fetching initial data: {str(e)}")
        states = []
    
    return render_template('index.html', 
                           states=states,
                           today=datetime.today().strftime('%Y-%m-%d'))

@app.route('/filter_data', methods=['POST'])
def filter_data():
    state = request.form.get('state')
    district = request.form.get('district')
    market = request.form.get('market')
    commodity = request.form.get('commodity')
    language = request.form.get('language', 'English')  # Default to English
    
    df = fetch_market_data(state, district, market, commodity)
    plots = generate_plots(df)
    # Pass market and commodity to get_ai_insights
    insights = get_ai_insights(df, state, district, market, commodity, language) if state and district and not df.empty else ""
    
    market_table_html = """
    <div class="table-responsive">
        <table class="table table-striped table-bordered">
            <thead>
                <tr>
                    <th>State</th>
                    <th>District</th>
                    <th>Market</th>
                    <th>Commodity</th>
                    <th>Variety</th>
                    <th>Grade</th>
                    <th>Arrival Date</th>
                    <th>Min Price</th>
                    <th>Max Price</th>
                    <th>Modal Price</th>
                </tr>
            </thead>
            <tbody>
    """
    
    for _, row in df.iterrows():
        market_table_html += f"""
            <tr>
                <td>{row['state']}</td>
                <td>{row['district']}</td>
                <td>{row['market']}</td>
                <td>{row['commodity']}</td>
                <td>{row['variety']}</td>
                <td>{row['grade']}</td>
                <td>{row['arrival_date']}</td>
                <td>₹{row['min_price']}</td>
                <td>₹{row['max_price']}</td>
                <td>₹{row['modal_price']}</td>
            </tr>
        """
    market_table_html += "</tbody></table></div>"

    cheapest_crops = df.sort_values('modal_price', ascending=True).head(5)
    cheapest_table_html = """
    <div class="table-responsive">
        <table class="table table-sm table-bordered">
            <thead>
                <tr>
                    <th>Commodity</th>
                    <th>Market</th>
                    <th>Modal Price</th>
                </tr>
            </thead>
            <tbody>
    """
    
    for _, row in cheapest_crops.iterrows():
        cheapest_table_html += f"""
            <tr>
                <td>{row['commodity']}</td>
                <td>{row['market']}</td>
                <td>₹{row['modal_price']}</td>
            </tr>
        """
    cheapest_table_html += "</tbody></table></div>"

    costliest_crops = df.sort_values('modal_price', ascending=False).head(5)
    costliest_table_html = """
    <div class="table-responsive">
        <table class="table table-sm table-bordered">
            <thead>
                <tr>
                    <th>Commodity</th>
                    <th>Market</th>
                    <th>Modal Price</th>
                </tr>
            </thead>
            <tbody>
    """
    
    for _, row in costliest_crops.iterrows():
        costliest_table_html += f"""
            <tr>
                <td>{row['commodity']}</td>
                <td>{row['market']}</td>
                <td>₹{row['modal_price']}</td>
            </tr>
        """
    costliest_table_html += "</tbody></table></div>"

    market_stats = {
        'total_commodities': len(df['commodity'].unique()),
        'avg_modal_price': f"₹{df['modal_price'].mean():.2f}",
        'price_range': f"₹{df['modal_price'].min():.2f} - ₹{df['modal_price'].max():.2f}",
        'total_markets': len(df['market'].unique())
    }

    response = {
        'plots': plots,
        'insights': insights,
        'success': not df.empty,
        'hasStateDistrict': bool(state and district),
        'market_html': market_table_html,
        'cheapest_html': cheapest_table_html,
        'costliest_html': costliest_table_html,
        'market_stats': market_stats
    }
    
    return jsonify(response)

@app.route('/get_districts', methods=['POST'])
def get_districts():
    state = request.form.get('state')
    df = fetch_market_data(state=state)
    districts = sorted(df['district'].dropna().unique())
    return jsonify(districts)

@app.route('/get_markets', methods=['POST'])
def get_markets():
    district = request.form.get('district')
    df = fetch_market_data(district=district)
    markets = sorted(df['market'].dropna().unique())
    return jsonify(markets)

@app.route('/get_commodities', methods=['POST'])
def get_commodities():
    market = request.form.get('market')
    df = fetch_market_data(market=market)
    commodities = sorted(df['commodity'].dropna().unique())
    return jsonify(commodities)

@app.route('/static/audio/<filename>')
def serve_audio(filename):
    try:
        return send_file(f"static/audio/{filename}", mimetype="audio/mpeg")
    except Exception as e:
        print(f"Error serving audio file: {str(e)}")
        return "Audio file not found", 404

if __name__ == '__main__':
    pio.templates.default = "plotly_white"
    app.run(debug=True, host='0.0.0.0', port=7860)