app.py

import streamlit as st
import pandas as pd
from fpdf import FPDF
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain.prompts import PromptTemplate
from dotenv import load_dotenv
import os

load_dotenv()

# --- CONFIGURATION ---
st.set_page_config(page_title="AdPulse | Automated Insights", page_icon="📊", layout="centered")

# --- Gemini API KEY from ENV ---
api_key = os.getenv("GEMINI_API_KEY")

# --- CORE LOGIC (PANDAS) ---
def analyze_campaign_data(df):
    """
    Performs deterministic calculations on the raw data.
    """
    try:
        # 1. Clean & Validate
        required_cols = ['Impressions', 'Clicks', 'Spend', 'Conversions']
        if not all(col in df.columns for col in required_cols):
            return None, "Missing required columns. Please upload a valid AdTech CSV."

        # 2. Aggregations
        total_spend = df['Spend'].sum()
        total_clicks = df['Clicks'].sum()
        total_conversions = df['Conversions'].sum()
        total_impressions = df['Impressions'].sum()

        # 3. Derived Metrics (KPIs)
        ctr = (total_clicks / total_impressions * 100) if total_impressions > 0 else 0
        cpa = total_spend / total_conversions if total_conversions > 0 else 0
        cpc = total_spend / total_clicks if total_clicks > 0 else 0
        
        # 4. Identify Winners & Losers
        df['ROAS_Proxy'] = df['Conversions'] / df['Spend'] # Simple Return on Ad Spend proxy
        top_campaign = df.loc[df['Conversions'].idxmax()]['Campaign_Name']
        most_efficient = df.loc[df['ROAS_Proxy'].idxmax()]['Platform']

        stats = {
            "Total Spend": f"${total_spend:,.2f}",
            "Total Conversions": f"{total_conversions:,}",
            "Global CTR": f"{ctr:.2f}%",
            "Avg CPA": f"${cpa:.2f}",
            "Avg CPC": f"${cpc:.2f}",
            "Top Campaign": top_campaign,
            "Best Platform": most_efficient
        }
        return stats, None

    except Exception as e:
        return None, f"Data Processing Error: {str(e)}"

# --- AI INSIGHT GENERATOR ---
def generate_executive_summary(stats, df_head):
    """
    Basically google gemini api is used to generate insights of the csv data.
    """
    if not api_key:
        return "⚠️ GEMINI_API_KEY is missing. Please check your .env file."

    try:
        # Using gemini-2.5-flash for speed and efficiency in the project
        llm = ChatGoogleGenerativeAI(
            model="gemini-2.5-flash",
            temperature=0.5,
            google_api_key=api_key,
            convert_system_message_to_human=True # act as a helper for some langchain versions, good to have
        )
        
        template = """
        You are a Senior Data Analyst at a top AdTech firm. 
        Write a concise, 3-paragraph executive summary for the Marketing Director based on the following weekly performance data.

        KEY METRICS:
        {metrics}

        CONTEXT:
        The highest performing platform is {best_platform}.
        
        INSTRUCTIONS:
        1. Start with an overall performance assessment.
        2. Highlight the 'Top Campaign' and why it succeeded.
        3. Provide one actionable recommendation for next week to lower the CPA.
        
        Keep the tone professional, objective, and action-oriented.
        """
        
        prompt = PromptTemplate(template=template, input_variables=["metrics", "best_platform"])
        
        # Format metrics for the prompt
        metrics_str = "\n".join([f"- {k}: {v}" for k,v in stats.items()])
        
        response = llm.predict(prompt.format(metrics=metrics_str, best_platform=stats['Best Platform']))
        return response
    
    except Exception as e:
        return f"AI Service Unavailable: {str(e)}"

# --- PDF GENERATION OF THE FINAL ANALYSIS---
class PDFReport(FPDF):
    def header(self):
        # 1. Top Dashboard Bar – GroundTruth style teal
        # Approx brand teal: rgb(0, 191, 165)
        self.set_fill_color(0, 191, 165)
        self.rect(0, 0, 210, 25, 'F')
        
        # 2. Logo / Title Area
        self.set_y(8)
        self.set_font('Arial', 'B', 16)
        self.set_text_color(255, 255, 255)  # White text
        self.cell(10)  # Left padding
        self.cell(0, 10, 'ADMINISTRATIVE DASHBOARD | WEEKLY PERFORMANCE', 0, 1, 'L')
        
        # 3. Sub-header (Date/Context)
        self.set_font('Arial', '', 9)
        # Slightly lighter text, still on teal
        self.set_text_color(230, 255, 248)
        self.cell(10)
        self.cell(0, 0, 'Generated by AdPulse AI Engine', 0, 1, 'L')
        self.ln(15)

    def footer(self):
        # Footer Bar – light mint background like website bg
        self.set_y(-15)
        # Soft mint / aqua strip
        self.set_fill_color(220, 247, 240)   # light mint
        self.rect(0, 282, 210, 15, 'F')
        self.set_font('Arial', 'I', 8)
        # Dark teal-ish grey text
        self.set_text_color(60, 90, 80)
        self.cell(0, 10, f'Confidential Internal Report | Page {self.page_no()}', 0, 0, 'C')


def create_pdf(stats, analysis_text):

    def safe_text(text):
        return text.encode('latin-1', 'replace').decode('latin-1')

    pdf = PDFReport()
    pdf.set_auto_page_break(auto=True, margin=15)
    pdf.add_page()
    
    # --- SECTION 1: KPI CARDS ---
    pdf.set_font("Arial", 'B', 12)
    # Dark teal for headings
    pdf.set_text_color(6, 64, 60)
    pdf.cell(0, 10, "1. HIGH-LEVEL METRICS", ln=True)
    pdf.ln(2)
    
    col_width = 60
    row_height = 25
    spacing = 5
    
    metrics_row_1 = [
        ("Total Spend", stats.get("Total Spend", "$0")),
        ("Conversions", stats.get("Total Conversions", "0")),
        ("Avg CPA", stats.get("Avg CPA", "$0"))
    ]
    
    metrics_row_2 = [
        ("Global CTR", stats.get("Global CTR", "0%")),
        ("Avg CPC", stats.get("Avg CPC", "$0")),
        ("Best Platform", stats.get("Best Platform", "N/A"))
    ]

    def draw_card(x, y, title, value):
        # Card border + fill with GroundTruth-like mint
        pdf.set_draw_color(210, 240, 232)      # mint border
        pdf.set_fill_color(242, 252, 249)      # very light mint fill
        pdf.rect(x, y, col_width, row_height, 'DF')
        
        # Card label (small, muted)
        pdf.set_xy(x + 2, y + 2)
        pdf.set_font("Arial", '', 8)
        pdf.set_text_color(90, 120, 110)       # grey-green label
        pdf.cell(col_width, 5, safe_text(title.upper()), 0, 0, 'L')
        
        # Card value (bold, dark teal)
        pdf.set_xy(x, y + 10)
        pdf.set_font("Arial", 'B', 14)
        pdf.set_text_color(6, 64, 60)
        if len(str(value)) > 15:
            pdf.set_font("Arial", 'B', 10)
        pdf.cell(col_width, 10, safe_text(str(value)), 0, 0, 'C')

    start_x = 10
    start_y = pdf.get_y()
    for i, (key, val) in enumerate(metrics_row_1):
        draw_card(start_x + (i * (col_width + spacing)), start_y, key, val)
    
    start_y += row_height + spacing
    for i, (key, val) in enumerate(metrics_row_2):
        draw_card(start_x + (i * (col_width + spacing)), start_y, key, val)

    pdf.set_y(start_y + row_height + 10)
    
    # --- SECTION 2: TOP CAMPAIGN ---
    # Light mint banner like hero section background
    pdf.set_fill_color(224, 248, 241)
    pdf.rect(10, pdf.get_y(), 190, 15, 'F')

    pdf.set_xy(15, pdf.get_y() + 4)
    pdf.set_font("Arial", 'B', 10)
    pdf.set_text_color(6, 64, 60)  # dark teal

    label = "TOP CAMPAIGN (WINNER):"
    label_width = pdf.get_string_width(label) + 2
    pdf.cell(label_width, 8, label, 0, 0)
    
    pdf.set_font("Arial", '', 10)
    pdf.cell(0, 8, safe_text(stats.get("Top Campaign", "N/A")), 0, 1)

    pdf.ln(10)

    # --- SECTION 3: AI EXECUTIVE SUMMARY ---
    pdf.set_font("Arial", 'B', 12)
    pdf.set_text_color(6, 64, 60)
    pdf.cell(0, 10, "2. AI STRATEGIC NARRATIVE", ln=True)
    
    start_text_y = pdf.get_y() + 5
    pdf.set_font("Arial", size=10)
    # Neutral dark text (like site body copy)
    pdf.set_text_color(55, 71, 79)
    pdf.set_xy(15, start_text_y)
    
    pdf.multi_cell(180, 6, safe_text(analysis_text))
    end_text_y = pdf.get_y()
    
    # Left accent line in brand teal
    pdf.set_line_width(1)
    pdf.set_draw_color(0, 191, 165)
    pdf.line(12, start_text_y, 12, end_text_y)
    
    return pdf.output(dest='S').encode('latin-1', 'ignore')


# --- MAIN APPLICATION UI ---
def main():
    st.title("📊 AdPulse: Automated Insight Engine")
    st.markdown("""
    **Transform raw campaign data into executive decisions in seconds.**
    Upload your weekly CSV to generate a PDF report with AI-driven analysis.
    """)

    with st.expander("ℹ️ NOTE FOR EVALUATORS (Click to Expand)", expanded=True):
        st.markdown("""
        **I have included multiple test scenarios in the `custom_csv/` folder to demonstrate AI adaptability:**
        1.  `data_tiktok_win.csv`: Shows AI identifying a high-volume, low-cost winner.
        2.  `data_b2b_win.csv`: Shows AI understanding high-CPA logic for B2B (LinkedIn).
        And many more so you can test the solution's reliability across contexts.
        
        *Please upload these files to see the narrative change dynamically.*
        """)
    
    # 1. File Upload
    uploaded_file = st.file_uploader("Upload CSV Data", type=['csv'])
    
    if not uploaded_file:
        st.info("👆 Please upload a CSV file to begin analysis.")

        # Optional: Add a link to sample data if you host it
        # st.markdown("[Download Sample CSV](https://github.com/...)")

        return
    
    # 2. Process Data
    if uploaded_file:
        df = pd.read_csv(uploaded_file)
        st.write("### 🔍 Data Preview")
        st.dataframe(df.head(5))
        
        if st.button("🚀 Generate Executive Report"):
            with st.spinner("Analyzing performance metrics with Gemini..."):
                # Step A: Numeric Analysis
                stats, error = analyze_campaign_data(df)
                
                if error:
                    st.error(error)
                    return

                # Step B: AI Narrative
                summary = generate_executive_summary(stats, df.head())
                
                # Step C: Display Results
                st.success("Analysis Complete!")
                
                # Metrics Display
                col1, col2, col3 = st.columns(3)
                col1.metric("Total Spend", stats['Total Spend'])
                col2.metric("Conversions", stats['Total Conversions'])
                col3.metric("CPA", stats['Avg CPA'])

                with st.expander("Read Executive Summary"): 
                    st.write(summary)

                pdf_bytes = create_pdf(stats, summary)
                
                st.download_button(
                    label="📥 Download Official PDF Report",
                    data=pdf_bytes,
                    file_name="Executive_Ad_Report.pdf",
                    mime="application/pdf"
                )

if __name__ == "__main__":
    main()