app.py

import gradio as gr
import joblib
import pandas as pd
import numpy as np
import os
import json
from PIL import Image
import google.generativeai as genai

# --- Gemini API Configuration ---
# It's recommended to set the GOOGLE_API_KEY as an environment variable for security.
try:
    GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
    if GOOGLE_API_KEY:
        genai.configure(api_key=GOOGLE_API_KEY)
    else:
        print("Warning: GOOGLE_API_KEY environment variable not found. The application will not be able to predict churn.")
except Exception as e:
    print(f"Error configuring Gemini API: {e}")

# --- Model and Asset Loading (Kept for reference, but not used in the new prediction function) ---

def load_models():
    """Loads the original ML model, encoders, and scaler from disk."""
    try:
        model = joblib.load('models/churn_model.pkl')
        encoders = joblib.load('models/label_encoders.pkl')
        scaler = joblib.load('models/scaler.pkl')
        return model, encoders, scaler
    except FileNotFoundError:
        print("Warning: Original model files not found in 'models/' directory.")
        return None, None, None
    except Exception as e:
        print(f"An unexpected error occurred while loading local models: {e}")
        return None, None, None

model, encoders, scaler = load_models()

def load_image(image_name):
    """Loads an image from the 'images' folder."""
    try:
        img_path = os.path.join("images", image_name)
        return Image.open(img_path) if os.path.exists(img_path) else None
    except Exception as e:
        print(f"Error loading image {image_name}: {e}")
        return None

# --- Feature Constants ---
REGIONS = ['North', 'South', 'East', 'West', 'Central']
PLAN_TYPES = ['Prepaid', 'Postpaid']
CONTRACT_TYPES = ['Month-to-month', 'One year', 'Two year']
COMPLAINT_STATUS = ['Open', 'Closed', 'Not Applicable']
PAYMENT_METHODS = ['Electronic check', 'Mailed check', 'Bank transfer', 'Credit card']

# --- Prediction Logic using Gemini API ---

def predict_churn(customer_id, region, plan_type, monthly_charges, total_charges,
                  tenure_months, contract_type, paperless_billing, payment_method,
                  data_usage_gb, call_minutes, sms_count, complaint_status, complaint_count):
    """Predicts customer churn using the Gemini API and generates a detailed result."""
    if not GOOGLE_API_KEY:
        return "🔴 **Error:** `GOOGLE_API_KEY` not configured. Please set it as an environment variable to enable predictions.", 0.0

    try:
        # 1. Construct a detailed prompt for the Gemini API
        prompt = f"""
        You are an expert telecom churn prediction analyst AI. Your analysis must be sharp, concise, and business-focused.
        
        Based on the following customer data, predict the churn probability.
        
        **Customer Data:**
        - **Customer ID:** {customer_id}
        - **Region:** {region}
        - **Plan Type:** {plan_type}
        - **Contract Type:** {contract_type}
        - **Tenure:** {tenure_months} months
        - **Monthly Charges:** ₹{monthly_charges}
        - **Total Charges:** ₹{total_charges}
        - **Data Usage:** {data_usage_gb} GB/month
        - **Call Minutes:** {call_minutes} mins/month
        - **SMS Count:** {sms_count} texts/month
        - **Payment Method:** {payment_method}
        - **Paperless Billing:** {'Yes' if paperless_billing else 'No'}
        - **Last Complaint Status:** {complaint_status}
        - **Total Complaint Count:** {complaint_count}

        **Analysis Guidelines:**
        - **High Risk Factors:** Month-to-month contracts, short tenure (especially < 12 months), high monthly charges relative to tenure, and any open complaints are strong indicators of churn.
        - **Low Risk Factors:** Long-term contracts (One year, Two year), long tenure, a high ratio of total charges to monthly charges, and no complaint history suggest a loyal customer.
        
        Return your response **only** as a single, valid JSON object with two keys:
        1. "churn_probability": a float number between 0.0 and 1.0.
        2. "analysis": a brief, professional, single-paragraph analysis explaining the key factors that influenced your prediction for this specific customer.

        **Example JSON output:**
        {{
          "churn_probability": 0.72,
          "analysis": "This customer is at a high risk of churning primarily due to their month-to-month contract, which offers little long-term commitment. Combined with a relatively short tenure of 12 months, they are susceptible to competitive offers. Proactive engagement with a long-term plan incentive is recommended."
        }}
        """

        # 2. Call the Gemini API
        gemini_model = genai.GenerativeModel('gemini-2.5-flash')
        response = gemini_model.generate_content(prompt)

        # 3. Parse the JSON response
        # Clean up potential markdown formatting from the response
        response_text = response.text.strip().replace("```json", "").replace("```", "")
        parsed_response = json.loads(response_text)
        churn_probability = float(parsed_response['churn_probability'])
        analysis = parsed_response['analysis']

        # 4. Determine risk level and format the final output
        if churn_probability > 0.7:
            risk_level = "🔴 HIGH RISK"
        elif churn_probability > 0.4:
            risk_level = "🟡 MEDIUM RISK"
        else:
            risk_level = "🟢 LOW RISK"

        result = f"""
        ### Prediction for Customer `{customer_id}`
        - **Churn Risk Level:** **{risk_level}**
        - **Probability of Churn:** **{churn_probability:.1%}**
        - **AI-Powered Analysis:** {analysis}
        """
        return result, churn_probability

    except json.JSONDecodeError as e:
        print(f"JSON Parsing Error: {e}")
        print(f"Gemini Response that caused the error: {response.text}")
        return "🔴 **Error:** Could not parse the analysis from the AI model. The response was not in the expected format.", 0.0
    except Exception as e:
        print(f"An unexpected error occurred during Gemini prediction: {e}")
        return "An unexpected error occurred while contacting the AI model. Please check the logs and try again.", 0.0


# --- Gradio UI ---
with gr.Blocks(title="Telecom Churn Prediction - BRBRAITT Group 5", theme=gr.themes.Soft()) as app:
    # Header and other UI elements... (No changes needed here, keeping it the same as before)
    gr.Markdown("""
    # 🔮 Telecom Churn Prediction System
    **TIRTC Course: Advance AI/ML Training (Nokia) | Institution: BRBRAITT, Jabalpur | Group 5**
    ---
    This AI-powered system predicts customer churn using Google's Gemini model for real-time analysis.
    """)

    with gr.Tabs():
        # Tab 1: Prediction Interface
        with gr.TabItem("🎯 Churn Prediction"):
            with gr.Row():
                with gr.Column(scale=2):
                    gr.Markdown("### Enter Customer Details")
                    with gr.Row():
                        customer_id = gr.Textbox(label="Customer ID", value="CUST-001")
                        region = gr.Dropdown(choices=REGIONS, label="Region", value="North")
                        plan_type = gr.Dropdown(choices=PLAN_TYPES, label="Plan Type", value="Postpaid")
                    with gr.Row():
                        contract_type = gr.Dropdown(choices=CONTRACT_TYPES, label="Contract Type", value="Month-to-month")
                        payment_method = gr.Dropdown(choices=PAYMENT_METHODS, label="Payment Method", value="Electronic check")
                        paperless_billing = gr.Checkbox(label="Paperless Billing", value=True)
                    gr.Markdown("#### Service Usage & Charges")
                    with gr.Row():
                        monthly_charges = gr.Number(label="Monthly Charges (₹)", value=1000)
                        total_charges = gr.Number(label="Total Charges (₹)", value=12000)
                        tenure_months = gr.Number(label="Tenure (Months)", value=12)
                    with gr.Row():
                        data_usage_gb = gr.Number(label="Data Usage (GB)", value=15)
                        call_minutes = gr.Number(label="Call Minutes", value=500)
                        sms_count = gr.Number(label="SMS Count", value=100)
                    gr.Markdown("#### Customer Complaints")
                    with gr.Row():
                        complaint_status = gr.Dropdown(choices=COMPLAINT_STATUS, label="Last Complaint Status", value="Not Applicable")
                        complaint_count = gr.Number(label="Total Complaint Count", value=0)
                    
                    predict_btn = gr.Button("🔮 Predict Churn Risk", variant="primary", size="lg")

                with gr.Column(scale=1):
                    gr.Markdown("### 📊 Prediction Result")
                    prediction_output = gr.Markdown(value="*Results will be displayed here...*")
                    probability_gauge = gr.Number(label="Churn Probability", value=0.0, interactive=False)

            predict_btn.click(
                fn=predict_churn,
                inputs=[customer_id, region, plan_type, monthly_charges, total_charges,
                        tenure_months, contract_type, paperless_billing, payment_method,
                        data_usage_gb, call_minutes, sms_count, complaint_status, complaint_count],
                outputs=[prediction_output, probability_gauge]
            )

        # Tab 2: Business Insights
        with gr.TabItem("💡 Business Insights"):
            gr.Markdown("### Key Findings & Recommendations")
            gr.Markdown("""
            #### 🎯 Model Performance
            - **Accuracy:** 90%
            - **AUC Score:** 0.95
            - **Best Algorithm:** Random Forest Classifier
            #### 💼 Business Impact
            - **Monthly Revenue at Risk:** Over ₹12,250.
            - **Potential Annual Loss:** Over ₹147,000.
            #### 🔴 Top Churn Drivers
            1.  **Contract Type:** `Month-to-month` customers are at highest risk.
            2.  **Tenure:** New customers (0-12 months) are most likely to churn.
            3.  **Complaints:** A single open complaint doubles the likelihood of churn.
            """)
        
        # Tab 3: Visualizations
        with gr.TabItem("📈 Visualizations"):
            gr.Markdown("### Data Analysis Dashboard")
            image_files = [
                ("churn_distribution.png", "Overall Churn Distribution"),
                ("churn_by_contract.png", "Churn by Contract Type"),
                ("revenue_vs_churn.png", "Revenue Impact Analysis"),
                ("complaints_analysis.png", "Complaints Impact on Churn"),
                ("correlation_matrix.png", "Feature Correlation Matrix"),
            ]
            for img_file, title in image_files:
                img = load_image(img_file)
                if img:
                    gr.Image(img, label=title, show_label=True)
                else:
                    gr.Markdown(f"*{title} - Image not available*")
        
        # Tab 4: About Project
        with gr.TabItem("ℹ️ About"):
            gr.Markdown("""
            ### 🎓 Academic Project Details
            - **Course:** TIRTC - Advance AI/ML Training (Nokia)
            - **Institution:** BRBRAITT, Jabalpur
            - **Team (Group 5):** Abhay Gupta, Jay Kumar, Kripanshu Gupta, Ruhy Namdeo
            - **Tech Stack:** Scikit-learn, Pandas, Gradio, Gemini, Hugging Face
            ---
            **🏆 Project Status:** Complete | **📅 Last Updated:** October 2025 | **🔢 Version:** 1.3.0
            """)
    
    gr.Markdown("--- \n © 2025 BRBRAITT Group 5 | TIRTC Advance AI/ML Training")

# Launch the app
if __name__ == "__main__":
    app.launch(share=True)