app.py

# app.py

import pandas as pd
import gradio as gr

# ---------------------------------------------------
# HELPERS
# ---------------------------------------------------

from helper.vintage_helpers import (
    create_booking_vintage
)

from helper.data_merger import (
    merge_acq_perf
)

# ---------------------------------------------------
# METRICS
# ---------------------------------------------------

from metrics.mix_metrics import (
    calculate_vintage_mix,
    calculate_limit_mix
)

# ---------------------------------------------------
# ANALYTICS
# ---------------------------------------------------

from analytics.performance_analysis import (
    generate_metric_view
)

from analytics.ai_assistant import (
    generate_ai_answer
)

from analytics.deep_dive_agentic import (
    run_deep_dive_analysis
)

# ---------------------------------------------------
# VISUALIZATIONS - VINTAGE CURVES
# ---------------------------------------------------

from visualizations.vintage_curves import (
    generate_delinquency_metric_chart,
    generate_multi_metric_comparison,
    generate_segment_delinquency_curve
)

# ---------------------------------------------------
# VISUALIZATIONS - SEGMENT RANKING
# ---------------------------------------------------

from visualizations.segment_ranking import (
    generate_segment_risk_heatmap,
    generate_segment_risk_ranking,
    generate_multi_category_risk_comparison,
    calculate_portfolio_risk_summary
)

# ---------------------------------------------------
# LOAD DATA
# ---------------------------------------------------

acq = pd.read_csv(
    "data/acquisition.csv"
)

perf = pd.read_csv(
    "data/performance.csv"
)

# ---------------------------------------------------
# CREATE BOOKING VINTAGE
# ---------------------------------------------------

acq = create_booking_vintage(
    acq,
    booking_date_col="booking_date"
)

# ---------------------------------------------------
# CREATE MASTER PERFORMANCE DATASET
# ---------------------------------------------------

master_df = merge_acq_perf(
    acq_df=acq,
    perf_df=perf
)

# ---------------------------------------------------
# ACQUISITION ANALYSIS
# ---------------------------------------------------

def run_acquisition_analysis(
    analysis_type,
    category
):

    # -----------------------------------------
    # PORTFOLIO MIX
    # -----------------------------------------

    if analysis_type == "Portfolio Mix":

        result = (
            acq.groupby(
                ["booking_vintage", category]
            )
            .agg(
                count=("account_id", "nunique"),
                balance=("credit_limit", "sum")
            )
            .reset_index()
        )

        vintage_total = (
            result.groupby("booking_vintage")["count"]
            .transform("sum")
        )

        result["rate"] = (
            result["count"] / vintage_total
        ) * 100

        result["rate"] = (
            result["rate"]
            .round(2)
        )

    # -----------------------------------------
    # CREDIT LINE CONCENTRATION
    # -----------------------------------------

    elif analysis_type == "Credit Line Concentration":

        result = (
            acq.groupby(
                ["booking_vintage", category]
            )
            .agg(
                count=("account_id", "nunique"),
                balance=("credit_limit", "sum")
            )
            .reset_index()
        )

        vintage_total = (
            result.groupby("booking_vintage")["balance"]
            .transform("sum")
        )

        result["rate"] = (
            result["balance"] / vintage_total
        ) * 100

        result["rate"] = (
            result["rate"]
            .round(2)
        )

    else:

        return pd.DataFrame()

    # -----------------------------------------
    # STANDARDIZED OUTPUT
    # -----------------------------------------

    result = result.rename(
        columns={
            "booking_vintage": "Vintage",
            category: "Category",
            "count": "Count",
            "balance": "Balance",
            "rate": "Rate"
        }
    )

    return result[
        [
            "Vintage",
            "Category",
            "Count",
            "Balance",
            "Rate"
        ]
    ]

# ---------------------------------------------------
# PERFORMANCE ANALYSIS
# ---------------------------------------------------

def run_performance_analysis(
    metric_name,
    view_level
):

    # -----------------------------------------
    # VIEW MAPPING
    # -----------------------------------------

    view_mapping = {

        "Overall": None,

        "Channel":
        "sourcing_channel",

        "FICO":
        "fico_band",

        "City Tier":
        "city_tier",

        "Occupation":
        "occupation_type"
    }

    group_col = view_mapping[
        view_level
    ]

    # -----------------------------------------
    # CALL ANALYTICS ENGINE
    # -----------------------------------------

    result = generate_metric_view(
        df=master_df,
        metric_name=metric_name,
        group_col=group_col
    )

    # -----------------------------------------
    # STANDARDIZE OUTPUT
    # -----------------------------------------

    if group_col is not None:

        result = result.rename(
            columns={
                group_col: "Category"
            }
        )

    else:

        result["Category"] = "Overall"

    # -----------------------------------------
    # IDENTIFY RATE COLUMN
    # -----------------------------------------

    rate_col = [
        col for col in result.columns
        if "rate" in col.lower()
    ][0]

    # -----------------------------------------
    # OUTPUT FORMAT
    # -----------------------------------------

    final_result = pd.DataFrame()

    final_result["Vintage"] = (
        result["booking_vintage"]
    )

    final_result["Category"] = (
        result["Category"]
    )

    final_result["Count"] = (
        result["total_accounts"]
    )

    final_result["Balance"] = (
        result["total_balance"]
    )

    final_result["Rate"] = (
        result[rate_col]
        .round(2)
    )

    return final_result

# ---------------------------------------------------
# VINTAGE CURVES ANALYSIS
# ---------------------------------------------------

def generate_vintage_curve_single(
    metric_name
):
    """Generate single vintage curve for a metric."""
    try:
        fig = generate_delinquency_metric_chart(
            df=master_df,
            metric_name=metric_name,
            chart_type="line"
        )
        return fig
    except Exception as e:
        return f"Error generating vintage curve: {str(e)}"


def generate_vintage_curves_comparison():
    """Generate comparison of all vintage curves."""
    try:
        fig = generate_multi_metric_comparison(
            df=master_df,
            metrics=["30+@3", "30+@6", "60+@6", "Yr1 NCL"]
        )
        return fig
    except Exception as e:
        return f"Error generating comparison: {str(e)}"


def generate_segmented_vintage_curve(
    metric_name,
    category
):
    """Generate vintage curve segmented by category."""
    try:
        fig = generate_segment_delinquency_curve(
            df=master_df,
            metric_name=metric_name,
            category=category
        )
        return fig
    except Exception as e:
        return f"Error generating segmented curve: {str(e)}"

# ---------------------------------------------------
# SEGMENT RANKING ANALYSIS
# ---------------------------------------------------

def generate_segment_risk_heatmap_chart():
    """Generate risk heatmap across all segments and metrics."""
    try:
        fig = generate_segment_risk_heatmap(
            df=master_df
        )
        return fig
    except Exception as e:
        return f"Error generating heatmap: {str(e)}"


def generate_high_risk_segments_ranking(
    metric_name,
    category
):
    """Generate ranking of high-risk segments."""
    try:
        fig = generate_segment_risk_ranking(
            df=master_df,
            metric_name=metric_name,
            category=category,
            top_n=10
        )
        return fig
    except Exception as e:
        return f"Error generating ranking: {str(e)}"


def generate_multi_category_comparison(
    metric_name
):
    """Generate risk comparison across all categories."""
    try:
        fig = generate_multi_category_risk_comparison(
            df=master_df,
            metric_name=metric_name
        )
        return fig
    except Exception as e:
        return f"Error generating comparison: {str(e)}"


def generate_portfolio_summary():
    """Generate portfolio risk summary."""
    try:
        summary_df = calculate_portfolio_risk_summary(
            df=master_df
        )
        return summary_df
    except Exception as e:
        return f"Error generating summary: {str(e)}"


def ask_ai_question(question, as_of_month, segment, history):
    if not question or str(question).strip() == "":
        return history or [], history or [], ""
    try:
        answer = generate_ai_answer(
            question=question,
            df=master_df,
            as_of_month=as_of_month,
            segment=segment if segment != "" else None
        )
    except Exception as exc:
        answer = f"AI error: {str(exc)}"

    history = history or []
    history.append({"role": "user", "content": question})
    history.append({"role": "assistant", "content": answer})
    return history, history, ""


def run_deep_dive_on_question(question):
    if not question or str(question).strip() == "":
        return "Please enter a question."
    try:
        result = run_deep_dive_analysis(question, acq, perf, master_df)
        
        # Check if analysis was successful
        if not result["success"]:
            return f"**Error in Deep Dive Analysis:**\n{result.get('error', 'Unknown error')}"
        
        # Check if we have empty results
        if not result.get("requirements") and not result.get("all_results"):
            return (
                f"**Question:** {result['question']}\n\n"
                f"⚠️ **No analysis requirements generated.** \n"
                f"This may indicate:\n"
                f"- The LLM couldn't parse the question clearly\n"
                f"- API connectivity issue\n"
                f"\nPlease try:\n"
                f"- Rephrasing your question more specifically\n"
                f"- Asking about specific metrics (30+@6, Yr1 NCL) or segments (FICO, Channel, City Tier)\n"
                f"- Breaking down complex questions into simpler components\n\n"
                f"**Raw Response:** {result.get('interpretation', 'No interpretation')}"
            )
        
        output = f"**Question:** {result['question']}\n\n"
        
        # Display each requirement and its result
        if result.get("requirements"):
            output += "## Analysis Requirements\n\n"
            for i, req in enumerate(result["requirements"], 1):
                output += f"**Requirement {i}:** {req.get('title', 'Untitled')}\n"
                output += f"- {req.get('description', 'No description')}\n\n"
                if "code" in req:
                    output += f"Generated Code:\n```python\n{req['code']}\n```\n\n"
        else:
            output += "⚠️ No requirements generated\n\n"
        
        # Display execution results
        if result.get("all_results"):
            output += "## Execution Results\n\n"
            for i, res in enumerate(result["all_results"], 1):
                output += f"**Requirement {i} Results:**\n"
                if res.get("success"):
                    output += f"✓ Executed successfully\n"
                    output += f"```\n{res.get('result', 'No result')}\n```\n\n"
                else:
                    output += f"✗ Execution failed: {res.get('error', 'Unknown error')}\n\n"
        else:
            output += "⚠️ No execution results\n\n"
        
        # Display synthesis & insights
        if result.get("interpretation"):
            output += "## Analyst Interpretation\n\n"
            output += f"{result['interpretation']}"
        else:
            output += "⚠️ No interpretation available"
        
        return output
    except Exception as exc:
        import traceback
        return f"**Deep Dive Analysis Error:**\n```\n{traceback.format_exc()}\n```"


# ---------------------------------------------------
# PORTFOLIO OVERVIEW (Calendar Snapshot)
# ---------------------------------------------------


def _detect_date_column(df):
    candidates = [
        "reporting_month",
        "observation_date",
        "observation_month",
        "obs_date",
        "date",
        "calendar_month",
        "month",
        "report_date"
    ]
    for c in candidates:
        if c in df.columns:
            return c
    return None


def get_calendar_months():
    date_col = _detect_date_column(master_df)
    if date_col is None:
        return []
    ser = pd.to_datetime(master_df[date_col], errors="coerce")
    months = ser.dt.to_period("M").astype(str).dropna().unique().tolist()
    months.sort()
    return months


def _filter_master_by_month(as_of_month):
    # as_of_month expected like "YYYY-MM"
    date_col = _detect_date_column(master_df)
    if date_col is None or not as_of_month:
        return master_df.copy()
    ser = pd.to_datetime(master_df[date_col], errors="coerce").dt.to_period("M").astype(str)
    return master_df[ser == as_of_month].copy()


def generate_portfolio_overview(as_of_month, segment):
    """
    Returns a small DataFrame with key portfolio snapshot metrics for the selected calendar month and segment.
    Metrics: Total Accounts, Open Accounts (balance>0), Bad Accounts (dpd>=30), Overall NCL Rate (dollar %), Average FICO.
    """
    df = _filter_master_by_month(as_of_month)

    # If a segmentation column is provided, return per-segment breakdown
    valid_segments = [
        "fico_band",
        "sourcing_channel",
        "city_tier",
        "occupation_type"
    ]

    if segment in valid_segments and segment in df.columns:
        grp = segment

        total_accounts = df.groupby(grp)["account_id"].nunique()

        if "balance" in df.columns:
            open_accounts = (
                df.loc[df["balance"] > 0].groupby(grp)["account_id"].nunique()
            )
            total_balance = df.groupby(grp)["balance"].sum()
        else:
            open_accounts = pd.Series(0, index=total_accounts.index)
            total_balance = pd.Series(0, index=total_accounts.index)

        if "dpd" in df.columns:
            bad_accounts = (
                df.loc[df["dpd"].fillna(0) >= 30].groupby(grp)["account_id"].nunique()
            )
            bad_balance = (
                df.assign(_bad_balance=df["balance"].where(df["dpd"].fillna(0) >= 30, 0))
                .groupby(grp)["_bad_balance"].sum()
            ) if "balance" in df.columns else pd.Series(0, index=total_accounts.index)
        else:
            bad_accounts = pd.Series(0, index=total_accounts.index)
            bad_balance = pd.Series(0, index=total_accounts.index)

        # NCL if present
        ncl_cols = [c for c in df.columns if "ncl" in c.lower()]
        if len(ncl_cols) > 0:
            total_ncl = df.groupby(grp)[ncl_cols[0]].sum()
            ncl_rate = (total_ncl / total_balance * 100).round(2).fillna(0)
        else:
            ncl_rate = (bad_balance / total_balance * 100).round(2).fillna(0)

        # average fico per group if available
        if "fico_score" in df.columns:
            avg_fico = df.groupby(grp)["fico_score"].mean().round(1)
        else:
            avg_fico = pd.Series(float("nan"), index=total_accounts.index)

        result = pd.DataFrame({
            "Segment": total_accounts.index,
            "Total_Accounts": total_accounts.values,
            "Open_Accounts": open_accounts.reindex(total_accounts.index).fillna(0).astype(int).values,
            "Bad_Accounts": bad_accounts.reindex(total_accounts.index).fillna(0).astype(int).values,
            "Total_Balance": total_balance.reindex(total_accounts.index).fillna(0).values,
            "NCL_Rate_pct": ncl_rate.reindex(total_accounts.index).fillna(0).values,
            "Avg_FICO": avg_fico.reindex(total_accounts.index).fillna(float("nan")).values
        })

        # Sort by NCL rate descending
        result = result.sort_values("NCL_Rate_pct", ascending=False).reset_index(drop=True)

        return result

    # Default: single-line overview
    total_accounts = df["account_id"].nunique() if "account_id" in df.columns else 0

    open_accounts = (
        df.loc[df["balance"] > 0, "account_id"].nunique()
        if "balance" in df.columns
        else total_accounts
    )

    bad_accounts = (
        df.loc[df["dpd"].fillna(0) >= 30, "account_id"].nunique()
        if "dpd" in df.columns
        else 0
    )

    # overall NCL rate (dollar-based) fallback logic
    ncl_cols = [c for c in df.columns if "ncl" in c.lower()]
    overall_ncl_rate = None
    if len(ncl_cols) > 0 and "balance" in df.columns:
        ncl_sum = df[ncl_cols[0]].sum(skipna=True)
        bal_sum = df["balance"].sum(skipna=True)
        overall_ncl_rate = (ncl_sum / bal_sum * 100) if bal_sum > 0 else None
    else:
        # fallback: use bad balance / total balance as proxy
        if "balance" in df.columns and "dpd" in df.columns:
            bad_bal = df.loc[df["dpd"].fillna(0) >= 30, "balance"].sum()
            bal_sum = df["balance"].sum()
            overall_ncl_rate = (bad_bal / bal_sum * 100) if bal_sum > 0 else None

    if overall_ncl_rate is None:
        overall_ncl_rate = float("nan")
    else:
        overall_ncl_rate = round(overall_ncl_rate, 2)

    # average fico
    avg_fico = None
    if "fico_score" in df.columns:
        avg_fico = round(df["fico_score"].dropna().mean(), 1)
    elif "fico_band" in df.columns:
        def band_mid(b):
            try:
                parts = b.split("-")
                return (int(parts[0]) + int(parts[1])) / 2
            except Exception:
                return None
        mid_vals = df["fico_band"].dropna().apply(band_mid).dropna()
        avg_fico = round(mid_vals.mean(), 1) if not mid_vals.empty else float("nan")
    else:
        avg_fico = float("nan")

    overview = pd.DataFrame({
        "Metric": [
            "As Of Month",
            "Total Accounts",
            "Open Accounts",
            "Bad Accounts (dpd>=30)",
            "Overall NCL Rate (%)",
            "Average FICO"
        ],
        "Value": [
            as_of_month if as_of_month else "All",
            int(total_accounts),
            int(open_accounts),
            int(bad_accounts),
            overall_ncl_rate,
            avg_fico
        ]
    })

    return overview

# ---------------------------------------------------
# DYNAMIC DROPDOWNS
# ---------------------------------------------------

def update_analysis_dropdown(
    dataset
):

    # -----------------------------------------
    # ACQUISITION
    # -----------------------------------------

    if dataset == "Acquisition":

        return gr.update(
            choices=[
                "Portfolio Mix",
                "Credit Line Concentration"
            ],
            value="Portfolio Mix"
        )

    # -----------------------------------------
    # PERFORMANCE
    # -----------------------------------------

    elif dataset == "Performance":

        return gr.update(
            choices=[
                "30+@3",
                "30+@6",
                "60+@6",
                "Yr1 NCL"
            ],
            value="30+@6"
        )


def update_category_dropdown(
    dataset
):

    # -----------------------------------------
    # ACQUISITION
    # -----------------------------------------

    if dataset == "Acquisition":

        return gr.update(
            choices=[
                "fico_band",
                "sourcing_channel",
                "city_tier",
                "occupation_type"
            ],
            value="fico_band"
        )

    # -----------------------------------------
    # PERFORMANCE
    # -----------------------------------------

    elif dataset == "Performance":

        return gr.update(
            choices=[
                "Overall",
                "Channel",
                "FICO",
                "City Tier",
                "Occupation"
            ],
            value="Overall"
        )

# ---------------------------------------------------
# MASTER ROUTER
# ---------------------------------------------------

def run_analysis(
    dataset,
    analysis,
    category
):

    # -----------------------------------------
    # ACQUISITION
    # -----------------------------------------

    if dataset == "Acquisition":

        return run_acquisition_analysis(
            analysis_type=analysis,
            category=category
        )

    # -----------------------------------------
    # PERFORMANCE
    # -----------------------------------------

    elif dataset == "Performance":

        return run_performance_analysis(
            metric_name=analysis,
            view_level=category
        )

    else:

        return pd.DataFrame()

# ---------------------------------------------------
# GRADIO UI
# ---------------------------------------------------

with gr.Blocks() as app:

    gr.Markdown(
        "# Risk Analytics Manager Agent - Phase 2"
    )

    with gr.Tabs():

        # =================================================
        # TAB 1: BASIC ANALYSIS (Phase 1)
        # =================================================

        with gr.TabItem("📊 Basic Analysis"):

            gr.Markdown(
                "## Phase 1: Acquisition & Performance Analysis"
            )

            with gr.Row():

                dataset_dropdown = gr.Dropdown(
                    choices=[
                        "Acquisition",
                        "Performance"
                    ],
                    value="Acquisition",
                    label="Dataset"
                )

                analysis_dropdown = gr.Dropdown(
                    choices=[
                        "Portfolio Mix",
                        "Credit Line Concentration"
                    ],
                    value="Portfolio Mix",
                    label="Analysis"
                )

                category_dropdown = gr.Dropdown(
                    choices=[
                        "fico_band",
                        "sourcing_channel",
                        "city_tier",
                        "occupation_type"
                    ],
                    value="fico_band",
                    label="Category / View"
                )

            # -----------------------------------------
            # DYNAMIC DROPDOWNS
            # -----------------------------------------

            dataset_dropdown.change(
                fn=update_analysis_dropdown,
                inputs=dataset_dropdown,
                outputs=analysis_dropdown
            )

            dataset_dropdown.change(
                fn=update_category_dropdown,
                inputs=dataset_dropdown,
                outputs=category_dropdown
            )

            # -----------------------------------------
            # RUN BUTTON
            # -----------------------------------------

            run_button = gr.Button(
                "Run Analysis",
                variant="primary"
            )

            output_table = gr.Dataframe()

            run_button.click(
                fn=run_analysis,
                inputs=[
                    dataset_dropdown,
                    analysis_dropdown,
                    category_dropdown
                ],
                outputs=output_table
            )

        # =================================================
        # TAB 2: VINTAGE CURVES (Phase 2)
        # =================================================

        with gr.TabItem("📈 Vintage Curves"):

            gr.Markdown(
                "## Phase 2: Vintage Delinquency Curves Analysis"
            )

            with gr.Row():

                metric_dropdown = gr.Dropdown(
                    choices=[
                        "30+@3",
                        "30+@6",
                        "60+@6",
                        "Yr1 NCL"
                    ],
                    value="30+@6",
                    label="Delinquency Metric"
                )

                vintage_chart_type = gr.Radio(
                    choices=["Single Metric", "All Metrics Comparison"],
                    value="Single Metric",
                    label="Chart Type"
                )

            def update_vintage_view(metric, chart_type):
                if chart_type == "Single Metric":
                    return generate_vintage_curve_single(metric)
                else:
                    return generate_vintage_curves_comparison()

            vintage_chart = gr.Plot(
                label="Vintage Curve"
            )

            gen_vintage_btn = gr.Button(
                "Generate Vintage Curve",
                variant="primary"
            )

            gen_vintage_btn.click(
                fn=update_vintage_view,
                inputs=[metric_dropdown, vintage_chart_type],
                outputs=vintage_chart
            )

            gr.Markdown(
                "### Segmented Vintage Curves"
            )

            with gr.Row():

                segment_metric = gr.Dropdown(
                    choices=[
                        "30+@3",
                        "30+@6",
                        "60+@6",
                        "Yr1 NCL"
                    ],
                    value="30+@6",
                    label="Metric"
                )

                segment_category = gr.Dropdown(
                    choices=[
                        "fico_band",
                        "sourcing_channel",
                        "city_tier",
                        "occupation_type"
                    ],
                    value="fico_band",
                    label="Category"
                )

            segmented_chart = gr.Plot(
                label="Segmented Vintage Curve"
            )

            gen_segment_btn = gr.Button(
                "Generate Segmented Curve",
                variant="primary"
            )

            gen_segment_btn.click(
                fn=generate_segmented_vintage_curve,
                inputs=[segment_metric, segment_category],
                outputs=segmented_chart
            )

        # =================================================
        # TAB 3: SEGMENT RANKING (Phase 2)
        # =================================================

        with gr.TabItem("⚠️ Segment Ranking"):

            gr.Markdown(
                "## Phase 2: High-Risk Segment Analysis"
            )

            # --------- HEATMAP SECTION ---------

            gr.Markdown(
                "### 🔥 Overall Risk Heatmap"
            )

            gr.Markdown(
                "Risk scores across all delinquency metrics and segments"
            )

            heatmap_chart = gr.Plot(
                label="Risk Heatmap"
            )

            gen_heatmap_btn = gr.Button(
                "Generate Risk Heatmap",
                variant="primary"
            )

            gen_heatmap_btn.click(
                fn=generate_segment_risk_heatmap_chart,
                outputs=heatmap_chart
            )

            gr.Markdown(
                "---"
            )

            # --------- HIGH-RISK RANKING SECTION ---------

            gr.Markdown(
                "### 📊 High-Risk Segments Ranking"
            )

            with gr.Row():

                ranking_metric = gr.Dropdown(
                    choices=[
                        "30+@3",
                        "30+@6",
                        "60+@6",
                        "Yr1 NCL"
                    ],
                    value="30+@6",
                    label="Metric"
                )

                ranking_category = gr.Dropdown(
                    choices=[
                        "fico_band",
                        "sourcing_channel",
                        "city_tier",
                        "occupation_type"
                    ],
                    value="fico_band",
                    label="Category"
                )

            ranking_chart = gr.Plot(
                label="High-Risk Segments"
            )

            gen_ranking_btn = gr.Button(
                "Generate Risk Ranking",
                variant="primary"
            )

            gen_ranking_btn.click(
                fn=generate_high_risk_segments_ranking,
                inputs=[ranking_metric, ranking_category],
                outputs=ranking_chart
            )

            gr.Markdown(
                "---"
            )

            # --------- MULTI-CATEGORY COMPARISON ---------

            gr.Markdown(
                "### 🔀 Cross-Category Risk Comparison"
            )

            comparison_metric = gr.Dropdown(
                choices=[
                    "30+@3",
                    "30+@6",
                    "60+@6",
                    "Yr1 NCL"
                ],
                value="30+@6",
                label="Metric"
            )

            comparison_chart = gr.Plot(
                label="Multi-Category Comparison"
            )

            gen_comparison_btn = gr.Button(
                "Generate Comparison",
                variant="primary"
            )

            gen_comparison_btn.click(
                fn=generate_multi_category_comparison,
                inputs=comparison_metric,
                outputs=comparison_chart
            )

            gr.Markdown(
                "---"
            )

            # --------- PORTFOLIO SUMMARY ---------

            gr.Markdown(
                "### 📋 Portfolio Risk Summary"
            )

            summary_table = gr.Dataframe(
                label="Risk Summary"
            )

            gen_summary_btn = gr.Button(
                "Generate Summary",
                variant="primary"
            )

            gen_summary_btn.click(
                fn=generate_portfolio_summary,
                outputs=summary_table
            )

        # =================================================
        # TAB 4: PORTFOLIO OVERVIEW (Calendar Snapshot)
        # =================================================

        with gr.TabItem("📅 Portfolio Overview"):

            gr.Markdown("## Portfolio Snapshot by Calendar Month")

            with gr.Row():
                calendar_month_dropdown = gr.Dropdown(
                    choices=get_calendar_months(),
                    value=(get_calendar_months()[-1] if len(get_calendar_months()) > 0 else None),
                    label="Calendar Month (YYYY-MM)"
                )

                overview_segment_dropdown = gr.Dropdown(
                    choices=[
                        "fico_band",
                        "sourcing_channel",
                        "city_tier",
                        "occupation_type"
                    ],
                    value="fico_band",
                    label="Segment (for drill)"
                )

                gen_overview_btn = gr.Button("Generate Snapshot", variant="primary")

            overview_table = gr.Dataframe(label="Portfolio Overview")

            gen_overview_btn.click(
                fn=generate_portfolio_overview,
                inputs=[calendar_month_dropdown, overview_segment_dropdown],
                outputs=overview_table
            )

        # =================================================
        # TAB 5: DEEP DIVE AGENTIC ANALYSIS
        # =================================================

        with gr.TabItem("🔬 Deep Dive Analysis"):

            gr.Markdown(
                "## Agentic Code Generation & Execution\n\n"
                "Ask a complex question and the AI will:\n"
                "1. Generate pandas code to analyze the data\n"
                "2. Execute the code safely\n"
                "3. Interpret the results with risk management insights"
            )

            deep_dive_question = gr.Textbox(
                lines=4,
                placeholder="E.g., 'Compare 30+ delinquency trends across fico bands and identify worst performing vintages.'",
                label="Analysis Question"
            )

            deep_dive_run_btn = gr.Button("Run Deep Dive Analysis", variant="primary")

            deep_dive_output = gr.Markdown(label="Analysis Output")

            deep_dive_run_btn.click(
                fn=run_deep_dive_on_question,
                inputs=deep_dive_question,
                outputs=deep_dive_output
            )

        # =================================================
        # TAB 6: CHAT WITH YOUR DATA
        # =================================================

        with gr.TabItem("💬 Chat with Your Data"):

            gr.Markdown(
                "## Ask questions in plain language and get risk manager insights"
            )

            with gr.Row():
                ai_month_dropdown = gr.Dropdown(
                    choices=["All"] + get_calendar_months(),
                    value="All" if len(get_calendar_months()) > 0 else None,
                    label="As Of Month (YYYY-MM or All)"
                )

                ai_segment_dropdown = gr.Dropdown(
                    choices=["", "fico_band", "sourcing_channel", "city_tier", "occupation_type"],
                    value="",
                    label="Segment Context"
                )

            ai_question = gr.Textbox(
                lines=3,
                placeholder="Ask about NCL, vintage performance, high-risk segments, or portfolio trends..."
            )

            ai_chatbot = gr.Chatbot()
            ai_history = gr.State([])

            ai_ask_btn = gr.Button("Ask AI", variant="primary")

            ai_ask_btn.click(
                fn=ask_ai_question,
                inputs=[ai_question, ai_month_dropdown, ai_segment_dropdown, ai_history],
                outputs=[ai_chatbot, ai_history, ai_question]
            )

app.launch()