app.py

# app.py
# (Showing relevant parts that need modification)

import gradio as gr
import pandas as pd
import os
import logging
import matplotlib
matplotlib.use('Agg') # Set backend for Matplotlib to avoid GUI conflicts with Gradio
import matplotlib.pyplot as plt
import time # For profiling if needed
from datetime import datetime, timedelta # Added timedelta
import numpy as np
from collections import OrderedDict # To maintain section order
import asyncio # For async operations with the new agent

# --- Module Imports ---
from gradio_utils import get_url_user_token
# Functions from newly created/refactored modules
from config import (
    LINKEDIN_CLIENT_ID_ENV_VAR, BUBBLE_APP_NAME_ENV_VAR,
    BUBBLE_API_KEY_PRIVATE_ENV_VAR, BUBBLE_API_ENDPOINT_ENV_VAR,
    PLOT_ID_TO_FORMULA_KEY_MAP)
from state_manager import process_and_store_bubble_token
from sync_logic import sync_all_linkedin_data_orchestrator
from ui_generators import (
    display_main_dashboard,
    run_mentions_tab_display,
    run_follower_stats_tab_display,
    build_analytics_tab_plot_area, # EXPECTED TO RETURN: plot_ui_objects, section_titles_map
    BOMB_ICON, EXPLORE_ICON, FORMULA_ICON, ACTIVE_ICON
)
from analytics_plot_generator import update_analytics_plots_figures, create_placeholder_plot
from formulas import PLOT_FORMULAS

# --- EXISTING CHATBOT MODULE IMPORTS ---
from chatbot_prompts import get_initial_insight_prompt_and_suggestions # MODIFIED IMPORT
from chatbot_handler import generate_llm_response
# --- END EXISTING CHATBOT MODULE IMPORTS ---

# --- NEW EMPLOYER BRANDING AGENT MODULE IMPORTS ---
try:
    from eb_agent_module import (
        EmployerBrandingAgent, 
        GENERATION_CONFIG_PARAMS as EB_AGENT_GEN_CONFIG, # Rename to avoid conflict
        LLM_MODEL_NAME as EB_AGENT_LLM_MODEL,           # Rename
        GEMINI_EMBEDDING_MODEL_NAME as EB_AGENT_EMBEDDING_MODEL, # Rename
        df_rag_documents as eb_agent_default_rag_docs, # Rename
        DEFAULT_SAFETY_SETTINGS as EB_AGENT_SAFETY_SETTINGS # Import safety settings
    )
    EB_AGENT_AVAILABLE = True
    logging.info("Successfully imported EmployerBrandingAgent module.")
except ImportError as e:
    logging.error(f"Failed to import EmployerBrandingAgent module: {e}", exc_info=True)
    EB_AGENT_AVAILABLE = False
    # Define dummy classes/variables if import fails, so app can still run
    class EmployerBrandingAgent:
        def __init__(self, *args, **kwargs): logging.error("EB Agent Dummy Class Initialized")
        async def process_query(self, query, **kwargs): return "# Error: Employer Branding Agent module not loaded."
        def update_dataframes(self, dfs): pass
        def clear_chat_history(self): pass
    EB_AGENT_GEN_CONFIG, EB_AGENT_LLM_MODEL, EB_AGENT_EMBEDDING_MODEL, eb_agent_default_rag_docs, EB_AGENT_SAFETY_SETTINGS = {}, None, None, pd.DataFrame(), {}


# Configure logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(module)s - %(message)s')


# --- Gradio UI Blocks ---
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky"),
               title="LinkedIn Organization Dashboard") as app:
    token_state = gr.State(value={
        "token": None, "client_id": None, "org_urn": None,
        "bubble_posts_df": pd.DataFrame(), "bubble_post_stats_df": pd.DataFrame(),
        "bubble_mentions_df": pd.DataFrame(), "bubble_follower_stats_df": pd.DataFrame(),
        "fetch_count_for_api": 0, "url_user_token_temp_storage": None,
        "config_date_col_posts": "published_at", "config_date_col_mentions": "date",
        "config_date_col_followers": "date", "config_media_type_col": "media_type",
        "config_eb_labels_col": "li_eb_label"
    })

    # States for existing analytics tab chatbot
    chat_histories_st = gr.State({})
    current_chat_plot_id_st = gr.State(None)
    plot_data_for_chatbot_st = gr.State({})

    # --- NEW: States for Employer Branding Agent Tab ---
    eb_agent_chat_history_st = gr.State([]) 
    # The agent instance itself will be created on-the-fly or managed if complex state is needed.
    # For now, we'll re-initialize it with fresh data in the handler.

    gr.Markdown("# 🚀 LinkedIn Organization Dashboard")
    url_user_token_display = gr.Textbox(label="User Token (Nascosto)", interactive=False, visible=False)
    status_box = gr.Textbox(label="Stato Generale Token LinkedIn", interactive=False, value="Inizializzazione...")
    org_urn_display = gr.Textbox(label="URN Organizzazione (Nascosto)", interactive=False, visible=False)

    app.load(fn=get_url_user_token, inputs=None, outputs=[url_user_token_display, org_urn_display], api_name="get_url_params", show_progress=False)

    def initial_load_sequence(url_token, org_urn_val, current_state):
        status_msg, new_state, btn_update = process_and_store_bubble_token(url_token, org_urn_val, current_state)
        dashboard_content = display_main_dashboard(new_state)
        return status_msg, new_state, btn_update, dashboard_content

    with gr.Tabs() as tabs:
        with gr.TabItem("1️⃣ Dashboard & Sync", id="tab_dashboard_sync"):
            gr.Markdown("Il sistema controlla i dati esistenti da Bubble. 'Sincronizza' si attiva se sono necessari nuovi dati.")
            sync_data_btn = gr.Button("🔄 Sincronizza Dati LinkedIn", variant="primary", visible=False, interactive=False)
            sync_status_html_output = gr.HTML("<p style='text-align:center;'>Stato sincronizzazione...</p>")
            dashboard_display_html = gr.HTML("<p style='text-align:center;'>Caricamento dashboard...</p>")

            org_urn_display.change(
                fn=initial_load_sequence,
                inputs=[url_user_token_display, org_urn_display, token_state],
                outputs=[status_box, token_state, sync_data_btn, dashboard_display_html],
                show_progress="full"
            )

        with gr.TabItem("2️⃣ Analisi", id="tab_analytics"):
            gr.Markdown("## 📈 Analisi Performance LinkedIn")
            gr.Markdown("Seleziona un intervallo di date. Clicca i pulsanti (💣 Insights, ƒ Formula, 🧭 Esplora) su un grafico per azioni.")
            analytics_status_md = gr.Markdown("Stato analisi...")

            with gr.Row():
                date_filter_selector = gr.Radio(
                    ["Sempre", "Ultimi 7 Giorni", "Ultimi 30 Giorni", "Intervallo Personalizzato"],
                    label="Seleziona Intervallo Date", value="Sempre", scale=3
                )
                with gr.Column(scale=2):
                    custom_start_date_picker = gr.DateTime(label="Data Inizio", visible=False, include_time=False, type="datetime")
                    custom_end_date_picker = gr.DateTime(label="Data Fine", visible=False, include_time=False, type="datetime")

            apply_filter_btn = gr.Button("🔍 Applica Filtro & Aggiorna Analisi", variant="primary")

            def toggle_custom_date_pickers(selection):
                is_custom = selection == "Intervallo Personalizzato"
                return gr.update(visible=is_custom), gr.update(visible=is_custom)

            date_filter_selector.change(
                fn=toggle_custom_date_pickers,
                inputs=[date_filter_selector],
                outputs=[custom_start_date_picker, custom_end_date_picker]
            )

            plot_configs = [
                {"label": "Numero di Follower nel Tempo", "id": "followers_count", "section": "Dinamiche dei Follower"},
                {"label": "Tasso di Crescita Follower", "id": "followers_growth_rate", "section": "Dinamiche dei Follower"},
                {"label": "Follower per Località", "id": "followers_by_location", "section": "Demografia Follower"},
                {"label": "Follower per Ruolo (Funzione)", "id": "followers_by_role", "section": "Demografia Follower"},
                {"label": "Follower per Settore", "id": "followers_by_industry", "section": "Demografia Follower"},
                {"label": "Follower per Anzianità", "id": "followers_by_seniority", "section": "Demografia Follower"},
                {"label": "Tasso di Engagement nel Tempo", "id": "engagement_rate", "section": "Approfondimenti Performance Post"},
                {"label": "Copertura nel Tempo", "id": "reach_over_time", "section": "Approfondimenti Performance Post"},
                {"label": "Visualizzazioni nel Tempo", "id": "impressions_over_time", "section": "Approfondimenti Performance Post"},
                {"label": "Reazioni (Like) nel Tempo", "id": "likes_over_time", "section": "Approfondimenti Performance Post"},
                {"label": "Click nel Tempo", "id": "clicks_over_time", "section": "Engagement Dettagliato Post nel Tempo"},
                {"label": "Condivisioni nel Tempo", "id": "shares_over_time", "section": "Engagement Dettagliato Post nel Tempo"},
                {"label": "Commenti nel Tempo", "id": "comments_over_time", "section": "Engagement Dettagliato Post nel Tempo"},
                {"label": "Ripartizione Commenti per Sentiment", "id": "comments_sentiment", "section": "Engagement Dettagliato Post nel Tempo"},
                {"label": "Frequenza Post", "id": "post_frequency_cs", "section": "Analisi Strategia Contenuti"},
                {"label": "Ripartizione Contenuti per Formato", "id": "content_format_breakdown_cs", "section": "Analisi Strategia Contenuti"},
                {"label": "Ripartizione Contenuti per Argomenti", "id": "content_topic_breakdown_cs", "section": "Analisi Strategia Contenuti"},
                {"label": "Volume Menzioni nel Tempo (Dettaglio)", "id": "mention_analysis_volume", "section": "Analisi Menzioni (Dettaglio)"},
                {"label": "Ripartizione Menzioni per Sentiment (Dettaglio)", "id": "mention_analysis_sentiment", "section": "Analisi Menzioni (Dettaglio)"}
            ]
            assert len(plot_configs) == 19, "Mancata corrispondenza in plot_configs e grafici attesi."
            
            unique_ordered_sections = list(OrderedDict.fromkeys(pc["section"] for pc in plot_configs))
            num_unique_sections = len(unique_ordered_sections)

            active_panel_action_state = gr.State(None) 
            explored_plot_id_state = gr.State(None) 

            plot_ui_objects = {} 
            section_titles_map = {} 

            with gr.Row(equal_height=False):
                with gr.Column(scale=8) as plots_area_col:
                    ui_elements_tuple = build_analytics_tab_plot_area(plot_configs)
                    if isinstance(ui_elements_tuple, tuple) and len(ui_elements_tuple) == 2:
                        plot_ui_objects, section_titles_map = ui_elements_tuple
                        if not all(sec_name in section_titles_map for sec_name in unique_ordered_sections):
                            logging.error("section_titles_map from build_analytics_tab_plot_area is incomplete.")
                            for sec_name in unique_ordered_sections:
                                if sec_name not in section_titles_map:
                                    section_titles_map[sec_name] = gr.Markdown(f"### {sec_name} (Error Placeholder)") 
                    else:
                        logging.error("build_analytics_tab_plot_area did not return a tuple of (plot_ui_objects, section_titles_map).")
                        plot_ui_objects = ui_elements_tuple if isinstance(ui_elements_tuple, dict) else {} 
                        for sec_name in unique_ordered_sections: 
                            section_titles_map[sec_name] = gr.Markdown(f"### {sec_name} (Error Placeholder)")


                with gr.Column(scale=4, visible=False) as global_actions_column_ui: 
                    gr.Markdown("### 💡 Azioni Contestuali Grafico")
                    insights_chatbot_ui = gr.Chatbot(
                        label="Chat Insights", type="messages", height=450,
                        bubble_full_width=False, visible=False, show_label=False,
                        placeholder="L'analisi AI del grafico apparirà qui. Fai domande di approfondimento!"
                    )
                    insights_chat_input_ui = gr.Textbox(
                        label="La tua domanda:", placeholder="Chiedi all'AI riguardo a questo grafico...",
                        lines=2, visible=False, show_label=False
                    )
                    with gr.Row(visible=False) as insights_suggestions_row_ui:
                        insights_suggestion_1_btn = gr.Button(value="Suggerimento 1", size="sm", min_width=50)
                        insights_suggestion_2_btn = gr.Button(value="Suggerimento 2", size="sm", min_width=50)
                        insights_suggestion_3_btn = gr.Button(value="Suggerimento 3", size="sm", min_width=50)

                    formula_display_markdown_ui = gr.Markdown(
                        "I dettagli sulla formula/metodologia appariranno qui.", visible=False
                    )
                    formula_close_hint_md = gr.Markdown( # Component for the hint's visibility
                        "<p style='font-size:0.9em; text-align:center; margin-top:10px;'><em>Click the active ƒ button on the plot again to close this panel.</em></p>", 
                        visible=False
                    )

            # --- ASYNC HANDLERS FOR ANALYTICS TAB ---
            async def handle_panel_action(
                plot_id_clicked: str, action_type: str, current_active_action_from_state: dict,
                current_chat_histories: dict, current_chat_plot_id: str,
                current_plot_data_for_chatbot: dict, current_explored_plot_id: str 
            ):
                logging.info(f"Panel Action: '{action_type}' for plot '{plot_id_clicked}'. Active: {current_active_action_from_state}, Explored: {current_explored_plot_id}")

                clicked_plot_config = next((p for p in plot_configs if p["id"] == plot_id_clicked), None)
                if not clicked_plot_config:
                    logging.error(f"Config not found for plot_id {plot_id_clicked}")
                    num_plots = len(plot_configs)
                    error_list_len = 15 + (4 * num_plots) + num_unique_sections 
                    error_list = [gr.update()] * error_list_len
                    error_list[11] = current_active_action_from_state 
                    error_list[12] = current_chat_plot_id 
                    error_list[13] = current_chat_histories 
                    error_list[14] = current_explored_plot_id 
                    return error_list

                clicked_plot_label = clicked_plot_config["label"]
                clicked_plot_section = clicked_plot_config["section"]
                hypothetical_new_active_state = {"plot_id": plot_id_clicked, "type": action_type}
                is_toggling_off = current_active_action_from_state == hypothetical_new_active_state

                action_col_visible_update = gr.update(visible=False)
                insights_chatbot_visible_update, insights_chat_input_visible_update, insights_suggestions_row_visible_update = gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
                formula_display_visible_update = gr.update(visible=False)
                formula_close_hint_visible_update = gr.update(visible=False) 
                chatbot_content_update, s1_upd, s2_upd, s3_upd, formula_content_update = gr.update(), gr.update(), gr.update(), gr.update(), gr.update()
                
                new_active_action_state_to_set, new_current_chat_plot_id = None, current_chat_plot_id
                updated_chat_histories, new_explored_plot_id_to_set = current_chat_histories, current_explored_plot_id

                generated_panel_vis_updates = []
                generated_bomb_btn_updates = []
                generated_formula_btn_updates = []
                generated_explore_btn_updates = []
                section_title_vis_updates = [gr.update()] * num_unique_sections


                if is_toggling_off:
                    new_active_action_state_to_set = None 
                    action_col_visible_update = gr.update(visible=False) 
                    logging.info(f"Toggling OFF panel {action_type} for {plot_id_clicked}.")

                    for _ in plot_configs:
                        generated_bomb_btn_updates.append(gr.update(value=BOMB_ICON))
                        generated_formula_btn_updates.append(gr.update(value=FORMULA_ICON))

                    if current_explored_plot_id: 
                        explored_cfg = next((p for p in plot_configs if p["id"] == current_explored_plot_id), None)
                        explored_sec = explored_cfg["section"] if explored_cfg else None
                        for i, sec_name in enumerate(unique_ordered_sections):
                            section_title_vis_updates[i] = gr.update(visible=(sec_name == explored_sec))
                        for cfg in plot_configs:
                            is_exp = (cfg["id"] == current_explored_plot_id)
                            generated_panel_vis_updates.append(gr.update(visible=is_exp))
                            generated_explore_btn_updates.append(gr.update(value=ACTIVE_ICON if is_exp else EXPLORE_ICON))
                    else: 
                        for i in range(num_unique_sections): section_title_vis_updates[i] = gr.update(visible=True)
                        for _ in plot_configs:
                            generated_panel_vis_updates.append(gr.update(visible=True))
                            generated_explore_btn_updates.append(gr.update(value=EXPLORE_ICON)) 
                    
                    if action_type == "insights": new_current_chat_plot_id = None

                else: # Toggling ON a new action or switching actions
                    new_active_action_state_to_set = hypothetical_new_active_state
                    action_col_visible_update = gr.update(visible=True) 
                    new_explored_plot_id_to_set = None 
                    logging.info(f"Toggling ON panel {action_type} for {plot_id_clicked}. Cancelling explore view if any.")

                    for i, sec_name in enumerate(unique_ordered_sections): 
                        section_title_vis_updates[i] = gr.update(visible=(sec_name == clicked_plot_section))
                    for cfg in plot_configs: 
                        generated_panel_vis_updates.append(gr.update(visible=(cfg["id"] == plot_id_clicked)))
                        generated_explore_btn_updates.append(gr.update(value=EXPLORE_ICON)) 

                    for cfg_btn in plot_configs: 
                        is_act_ins = new_active_action_state_to_set == {"plot_id": cfg_btn["id"], "type": "insights"}
                        is_act_for = new_active_action_state_to_set == {"plot_id": cfg_btn["id"], "type": "formula"}
                        generated_bomb_btn_updates.append(gr.update(value=ACTIVE_ICON if is_act_ins else BOMB_ICON))
                        generated_formula_btn_updates.append(gr.update(value=ACTIVE_ICON if is_act_for else FORMULA_ICON))
                    
                    if action_type == "insights":
                        insights_chatbot_visible_update, insights_chat_input_visible_update, insights_suggestions_row_visible_update = gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
                        new_current_chat_plot_id = plot_id_clicked
                        history = current_chat_histories.get(plot_id_clicked, [])
                        summary = current_plot_data_for_chatbot.get(plot_id_clicked, f"No summary for '{clicked_plot_label}'.")
                        if not history: 
                            prompt, sugg = get_initial_insight_prompt_and_suggestions(plot_id_clicked, clicked_plot_label, summary)
                            llm_hist = [{"role": "user", "content": prompt}]
                            resp = await generate_llm_response(prompt, plot_id_clicked, clicked_plot_label, llm_hist, summary) # This is your existing LLM call
                            history = [{"role": "assistant", "content": resp}] 
                            updated_chat_histories = {**current_chat_histories, plot_id_clicked: history}
                        else: 
                            _, sugg = get_initial_insight_prompt_and_suggestions(plot_id_clicked, clicked_plot_label, summary) 
                        
                        chatbot_content_update = gr.update(value=history)
                        s1_upd,s2_upd,s3_upd = gr.update(value=sugg[0] if sugg else "N/A"),gr.update(value=sugg[1] if len(sugg)>1 else "N/A"),gr.update(value=sugg[2] if len(sugg)>2 else "N/A")

                    elif action_type == "formula":
                        formula_display_visible_update = gr.update(visible=True)
                        formula_close_hint_visible_update = gr.update(visible=True) 
                        f_key = PLOT_ID_TO_FORMULA_KEY_MAP.get(plot_id_clicked)
                        f_text = f"**Formula/Methodology for: {clicked_plot_label}** (ID: `{plot_id_clicked}`)\n\n"
                        if f_key and f_key in PLOT_FORMULAS:
                            f_data = PLOT_FORMULAS[f_key]
                            f_text += f"### {f_data['title']}\n\n{f_data['description']}\n\n**Calculation:**\n" + "\n".join([f"- {s}" for s in f_data['calculation_steps']])
                        else: f_text += "(No detailed formula information found.)"
                        formula_content_update = gr.update(value=f_text)
                        new_current_chat_plot_id = None 

                final_updates = [
                    action_col_visible_update, insights_chatbot_visible_update, chatbot_content_update,
                    insights_chat_input_visible_update, insights_suggestions_row_visible_update, 
                    s1_upd, s2_upd, s3_upd, formula_display_visible_update, formula_content_update,
                    formula_close_hint_visible_update, # Corrected from formula_close_hint_md
                    new_active_action_state_to_set, new_current_chat_plot_id, updated_chat_histories,
                    new_explored_plot_id_to_set 
                ]

                final_updates.extend(generated_panel_vis_updates)   
                final_updates.extend(generated_bomb_btn_updates)    
                final_updates.extend(generated_formula_btn_updates) 
                final_updates.extend(generated_explore_btn_updates) 
                final_updates.extend(section_title_vis_updates)     
                
                logging.debug(f"handle_panel_action returning {len(final_updates)} updates. Expected {15 + 4*len(plot_configs) + num_unique_sections}.")
                return final_updates

            async def handle_chat_message_submission(user_message: str, current_plot_id: str, chat_histories: dict, current_plot_data_for_chatbot: dict ):
                if not current_plot_id or not user_message.strip():
                    current_history_for_plot = chat_histories.get(current_plot_id, [])
                    if not isinstance(current_history_for_plot, list): current_history_for_plot = []
                    yield current_history_for_plot, gr.update(value=""), chat_histories; return
                
                cfg = next((p for p in plot_configs if p["id"] == current_plot_id), None)
                lbl = cfg["label"] if cfg else "Selected Plot"
                summary = current_plot_data_for_chatbot.get(current_plot_id, f"No summary for '{lbl}'.")
                
                hist_for_plot = chat_histories.get(current_plot_id, [])
                if not isinstance(hist_for_plot, list): hist_for_plot = [] 

                hist = hist_for_plot.copy() + [{"role": "user", "content": user_message}]
                yield hist, gr.update(value=""), chat_histories 

                resp = await generate_llm_response(user_message, current_plot_id, lbl, hist, summary) # Existing LLM
                hist.append({"role": "assistant", "content": resp})
                
                updated_chat_histories = {**chat_histories, current_plot_id: hist}
                yield hist, "", updated_chat_histories


            async def handle_suggested_question_click(suggestion_text: str, current_plot_id: str, chat_histories: dict, current_plot_data_for_chatbot: dict):
                if not current_plot_id or not suggestion_text.strip() or suggestion_text == "N/A":
                    current_history_for_plot = chat_histories.get(current_plot_id, [])
                    if not isinstance(current_history_for_plot, list): current_history_for_plot = []
                    yield current_history_for_plot, gr.update(value=""), chat_histories; return
                
                async for update_chunk in handle_chat_message_submission(suggestion_text, current_plot_id, chat_histories, current_plot_data_for_chatbot):
                    yield update_chunk


            def handle_explore_click(plot_id_clicked, current_explored_plot_id_from_state, current_active_panel_action_state):
                # This function remains synchronous as per original
                logging.info(f"Explore Click: Plot '{plot_id_clicked}'. Current Explored: {current_explored_plot_id_from_state}. Active Panel: {current_active_panel_action_state}")
                num_plots = len(plot_configs)
                if not plot_ui_objects: 
                    logging.error("plot_ui_objects not populated for handle_explore_click.")
                    error_list_len = 4 + (4 * num_plots) + num_unique_sections 
                    error_list = [gr.update()] * error_list_len
                    error_list[0] = current_explored_plot_id_from_state 
                    error_list[2] = current_active_panel_action_state 
                    return error_list

                new_explored_id_to_set = None
                is_toggling_off_explore = (plot_id_clicked == current_explored_plot_id_from_state)

                action_col_upd = gr.update() 
                new_active_panel_state_upd = current_active_panel_action_state 
                formula_hint_upd = gr.update(visible=False) 

                panel_vis_updates = []
                explore_btns_updates = []
                bomb_btns_updates = []
                formula_btns_updates = []
                section_title_vis_updates = [gr.update()] * num_unique_sections

                clicked_cfg = next((p for p in plot_configs if p["id"] == plot_id_clicked), None)
                sec_of_clicked = clicked_cfg["section"] if clicked_cfg else None

                if is_toggling_off_explore:
                    new_explored_id_to_set = None 
                    logging.info(f"Stopping explore for {plot_id_clicked}. All plots/sections to be visible.")
                    for i in range(num_unique_sections): section_title_vis_updates[i] = gr.update(visible=True)
                    for _ in plot_configs:
                        panel_vis_updates.append(gr.update(visible=True))
                        explore_btns_updates.append(gr.update(value=EXPLORE_ICON))
                        bomb_btns_updates.append(gr.update()) 
                        formula_btns_updates.append(gr.update()) 
                else: 
                    new_explored_id_to_set = plot_id_clicked
                    logging.info(f"Exploring {plot_id_clicked}. Hiding other plots/sections.")
                    for i, sec_name in enumerate(unique_ordered_sections): 
                        section_title_vis_updates[i] = gr.update(visible=(sec_name == sec_of_clicked))
                    for cfg in plot_configs:
                        is_target = (cfg["id"] == new_explored_id_to_set)
                        panel_vis_updates.append(gr.update(visible=is_target))
                        explore_btns_updates.append(gr.update(value=ACTIVE_ICON if is_target else EXPLORE_ICON))
                    
                    if current_active_panel_action_state:
                        logging.info("Closing active insight/formula panel due to explore click.")
                        action_col_upd = gr.update(visible=False) 
                        new_active_panel_state_upd = None 
                        formula_hint_upd = gr.update(visible=False) 
                        for _ in plot_configs:
                            bomb_btns_updates.append(gr.update(value=BOMB_ICON))
                            formula_btns_updates.append(gr.update(value=FORMULA_ICON))
                    else:
                        for _ in plot_configs:
                            bomb_btns_updates.append(gr.update())
                            formula_btns_updates.append(gr.update())
                
                final_explore_updates = [
                    new_explored_id_to_set, action_col_upd, new_active_panel_state_upd, formula_hint_upd
                ]
                final_explore_updates.extend(panel_vis_updates)
                final_explore_updates.extend(explore_btns_updates)
                final_explore_updates.extend(bomb_btns_updates)
                final_explore_updates.extend(formula_btns_updates)
                final_explore_updates.extend(section_title_vis_updates)
                
                logging.debug(f"handle_explore_click returning {len(final_explore_updates)} updates. Expected {4 + 4*len(plot_configs) + num_unique_sections}.")
                return final_explore_updates


            _base_action_panel_ui_outputs = [
                global_actions_column_ui, insights_chatbot_ui, insights_chatbot_ui, 
                insights_chat_input_ui, insights_suggestions_row_ui, 
                insights_suggestion_1_btn, insights_suggestion_2_btn, insights_suggestion_3_btn,
                formula_display_markdown_ui, formula_display_markdown_ui, 
                formula_close_hint_md 
            ]
            _action_panel_state_outputs = [active_panel_action_state, current_chat_plot_id_st, chat_histories_st, explored_plot_id_state]
            
            action_panel_outputs_list = _base_action_panel_ui_outputs + _action_panel_state_outputs 
            action_panel_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("panel_component", gr.update()) for pc in plot_configs]) 
            action_panel_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("bomb_button", gr.update()) for pc in plot_configs])   
            action_panel_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("formula_button", gr.update()) for pc in plot_configs]) 
            action_panel_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("explore_button", gr.update()) for pc in plot_configs]) 
            action_panel_outputs_list.extend([section_titles_map.get(s_name, gr.update()) for s_name in unique_ordered_sections]) 

            _explore_base_outputs = [explored_plot_id_state, global_actions_column_ui, active_panel_action_state, formula_close_hint_md] 
            explore_outputs_list = _explore_base_outputs 
            explore_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("panel_component", gr.update()) for pc in plot_configs]) 
            explore_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("explore_button", gr.update()) for pc in plot_configs]) 
            explore_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("bomb_button", gr.update()) for pc in plot_configs])   
            explore_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("formula_button", gr.update()) for pc in plot_configs]) 
            explore_outputs_list.extend([section_titles_map.get(s_name, gr.update()) for s_name in unique_ordered_sections]) 

            action_click_inputs = [active_panel_action_state, chat_histories_st, current_chat_plot_id_st, plot_data_for_chatbot_st, explored_plot_id_state]
            explore_click_inputs = [explored_plot_id_state, active_panel_action_state]

            def create_panel_action_handler(p_id, action_type_str):
                async def _handler(curr_active_val, curr_chats_val, curr_chat_pid, curr_plot_data, curr_explored_id):
                    return await handle_panel_action(p_id, action_type_str, curr_active_val, curr_chats_val, curr_chat_pid, curr_plot_data, curr_explored_id)
                return _handler

            for config_item in plot_configs:
                plot_id = config_item["id"]
                if plot_id in plot_ui_objects:
                    ui_obj = plot_ui_objects[plot_id]
                    if ui_obj.get("bomb_button"):
                        ui_obj["bomb_button"].click(fn=create_panel_action_handler(plot_id, "insights"), inputs=action_click_inputs, outputs=action_panel_outputs_list, api_name=f"action_insights_{plot_id}")
                    if ui_obj.get("formula_button"):
                        ui_obj["formula_button"].click(fn=create_panel_action_handler(plot_id, "formula"), inputs=action_click_inputs, outputs=action_panel_outputs_list, api_name=f"action_formula_{plot_id}")
                    if ui_obj.get("explore_button"):
                        # Original lambda was not async, ensure it matches handle_explore_click signature and type
                        ui_obj["explore_button"].click(
                            fn=lambda current_explored_val, current_active_panel_val, p_id=plot_id: handle_explore_click(p_id, current_explored_val, current_active_panel_val), 
                            inputs=explore_click_inputs, 
                            outputs=explore_outputs_list, 
                            api_name=f"action_explore_{plot_id}"
                        ) # if handle_explore_click becomes async, this needs 'await' or be wrapped
                else: logging.warning(f"UI object for plot_id '{plot_id}' not found for click handlers.")


            chat_submission_outputs = [insights_chatbot_ui, insights_chat_input_ui, chat_histories_st]
            chat_submission_inputs = [insights_chat_input_ui, current_chat_plot_id_st, chat_histories_st, plot_data_for_chatbot_st]
            insights_chat_input_ui.submit(fn=handle_chat_message_submission, inputs=chat_submission_inputs, outputs=chat_submission_outputs, api_name="submit_chat_message")
            
            suggestion_click_inputs_base = [current_chat_plot_id_st, chat_histories_st, plot_data_for_chatbot_st]
            insights_suggestion_1_btn.click(fn=handle_suggested_question_click, inputs=[insights_suggestion_1_btn] + suggestion_click_inputs_base, outputs=chat_submission_outputs, api_name="click_suggestion_1")
            insights_suggestion_2_btn.click(fn=handle_suggested_question_click, inputs=[insights_suggestion_2_btn] + suggestion_click_inputs_base, outputs=chat_submission_outputs, api_name="click_suggestion_2")
            insights_suggestion_3_btn.click(fn=handle_suggested_question_click, inputs=[insights_suggestion_3_btn] + suggestion_click_inputs_base, outputs=chat_submission_outputs, api_name="click_suggestion_3")


            def refresh_all_analytics_ui_elements(current_token_state_val, date_filter_val, custom_start_val, custom_end_val, current_chat_histories_val):
                # This function remains synchronous as per original
                logging.info("Refreshing all analytics UI elements and resetting actions/chat.")
                plot_gen_results = update_analytics_plots_figures(current_token_state_val, date_filter_val, custom_start_val, custom_end_val, plot_configs)
                status_msg, gen_figs, new_summaries = plot_gen_results[0], plot_gen_results[1:-1], plot_gen_results[-1]
                
                all_updates = [status_msg] 
                all_updates.extend(gen_figs if len(gen_figs) == len(plot_configs) else [create_placeholder_plot("Error", f"Fig missing {i}") for i in range(len(plot_configs))]) 
                
                all_updates.extend([ 
                    gr.update(visible=False), 
                    gr.update(value=[], visible=False), 
                    gr.update(value="", visible=False), 
                    gr.update(visible=False), 
                    gr.update(value="S1"), gr.update(value="S2"), gr.update(value="S3"), 
                    gr.update(value="Formula details here.", visible=False), 
                    gr.update(visible=False) 
                ]) 
                
                all_updates.extend([ 
                    None, 
                    None, 
                    {},   
                    new_summaries 
                ]) 
                
                for _ in plot_configs: 
                    all_updates.extend([
                        gr.update(value=BOMB_ICON), 
                        gr.update(value=FORMULA_ICON), 
                        gr.update(value=EXPLORE_ICON), 
                        gr.update(visible=True) 
                    ]) 
                
                all_updates.append(None) 
                all_updates.extend([gr.update(visible=True)] * num_unique_sections) 
                
                logging.info(f"Prepared {len(all_updates)} updates for analytics refresh. Expected {15 + 5*len(plot_configs) + num_unique_sections}.")
                return all_updates

            apply_filter_and_sync_outputs_list = [analytics_status_md] 
            apply_filter_and_sync_outputs_list.extend([plot_ui_objects.get(pc["id"], {}).get("plot_component", gr.update()) for pc in plot_configs]) 
            
            _ui_resets_for_filter = [
                global_actions_column_ui, insights_chatbot_ui, insights_chat_input_ui,
                insights_suggestions_row_ui, insights_suggestion_1_btn, insights_suggestion_2_btn, insights_suggestion_3_btn,
                formula_display_markdown_ui, formula_close_hint_md 
            ]
            apply_filter_and_sync_outputs_list.extend(_ui_resets_for_filter)
            
            _state_resets_for_filter = [active_panel_action_state, current_chat_plot_id_st, chat_histories_st, plot_data_for_chatbot_st]
            apply_filter_and_sync_outputs_list.extend(_state_resets_for_filter)

            for pc in plot_configs: 
                pid = pc["id"]
                apply_filter_and_sync_outputs_list.extend([
                    plot_ui_objects.get(pid, {}).get("bomb_button", gr.update()),
                    plot_ui_objects.get(pid, {}).get("formula_button", gr.update()),
                    plot_ui_objects.get(pid, {}).get("explore_button", gr.update()),
                    plot_ui_objects.get(pid, {}).get("panel_component", gr.update()) 
                ])
            
            apply_filter_and_sync_outputs_list.append(explored_plot_id_state) 
            
            apply_filter_and_sync_outputs_list.extend([section_titles_map.get(s_name, gr.update()) for s_name in unique_ordered_sections]) 
            
            apply_filter_btn.click(
                fn=refresh_all_analytics_ui_elements,
                inputs=[token_state, date_filter_selector, custom_start_date_picker, custom_end_date_picker, chat_histories_st],
                outputs=apply_filter_and_sync_outputs_list, show_progress="full"
            )

        with gr.TabItem("3️⃣ Menzioni", id="tab_mentions"):
            refresh_mentions_display_btn = gr.Button("🔄 Aggiorna Visualizzazione Menzioni", variant="secondary")
            mentions_html = gr.HTML("Dati menzioni...")
            mentions_sentiment_dist_plot = gr.Plot(label="Distribuzione Sentiment Menzioni")
            refresh_mentions_display_btn.click(
                fn=run_mentions_tab_display, inputs=[token_state],
                outputs=[mentions_html, mentions_sentiment_dist_plot],
                show_progress="full"
            )

        with gr.TabItem("4️⃣ Statistiche Follower", id="tab_follower_stats"):
            refresh_follower_stats_btn = gr.Button("🔄 Aggiorna Visualizzazione Statistiche Follower", variant="secondary")
            follower_stats_html = gr.HTML("Statistiche follower...")
            with gr.Row():
                fs_plot_monthly_gains = gr.Plot(label="Guadagni Mensili Follower")
            with gr.Row():
                fs_plot_seniority = gr.Plot(label="Follower per Anzianità (Top 10 Organici)")
                fs_plot_industry = gr.Plot(label="Follower per Settore (Top 10 Organici)")
            
            refresh_follower_stats_btn.click(
                fn=run_follower_stats_tab_display, inputs=[token_state],
                outputs=[follower_stats_html, fs_plot_monthly_gains, fs_plot_seniority, fs_plot_industry],
                show_progress="full"
            )
        
        # --- NEW: Tab 5 for Employer Branding Agent ---
        with gr.TabItem("5️⃣ Agente AI Employer Branding", id="tab_eb_agent"):
            gr.Markdown("## 🤖 Interagisci con l'Agente AI per l'Employer Branding")
            
            if not EB_AGENT_AVAILABLE:
                gr.Markdown("<p style='color:red;font-weight:bold;'>Attenzione: Il modulo dell'Agente AI per l'Employer Branding non è stato caricato correttamente. Controllare i log e l'installazione della libreria `google-generativeai` e la variabile d'ambiente `GEMINI_API_KEY`.</p>")
            elif not os.getenv('GEMINI_API_KEY'):
                 gr.Markdown("<p style='color:orange;font-weight:bold;'>Attenzione: La variabile d'ambiente `GEMINI_API_KEY` non è impostata. Le funzionalità dell'Agente AI saranno limitate o non funzioneranno.</p>")


            gr.Markdown(
                "Fai domande sui tuoi dati LinkedIn (statistiche follower, post e menzioni) per ottenere insights e codice Pandas per analizzarli. "
                "L'agente utilizza i dati attualmente disponibili nello stato dell'applicazione."
            )
            with gr.Row():
                with gr.Column(scale=2):
                    eb_agent_chatbot_ui = gr.Chatbot(
                        label="Chat con Agente AI EB",
                        value=[[None, "Ciao! Sono il tuo Agente AI per l'Employer Branding. Come posso aiutarti ad analizzare i tuoi dati LinkedIn oggi? Chiedimi di generare codice Pandas o di fornire insights."]] if EB_AGENT_AVAILABLE else [[None, "Agente AI non disponibile."]],
                        bubble_full_width=False,
                        height=500,
                        placeholder="L'Agente AI è pronto. Chiedi pure..."
                    )
                    eb_agent_chat_input_ui = gr.Textbox(
                        label="La tua domanda:",
                        placeholder="Es: 'Mostrami le aziende dei miei follower nel settore tecnologico' o 'Qual è il sentiment medio delle mie menzioni?'",
                        lines=3,
                        interactive=EB_AGENT_AVAILABLE # Disable if agent not available
                    )
                    with gr.Row():
                        eb_agent_submit_btn = gr.Button("💬 Invia Messaggio", variant="primary", interactive=EB_AGENT_AVAILABLE)
                        eb_agent_clear_btn = gr.Button("🗑️ Cancella Chat", variant="stop", interactive=EB_AGENT_AVAILABLE)
                with gr.Column(scale=1):
                    gr.Markdown("#### Schemi Dati Disponibili per l'Agente:")
                    eb_agent_schema_display_md = gr.Markdown("Gli schemi dei dati (follower, post, menzioni) verranno mostrati qui quando l'agente viene inizializzato con una query.")
                    eb_agent_status_md = gr.Markdown("Stato Agente: In attesa di input...")
            
            # --- NEW: Handler for Employer Branding Agent Chat ---
            eb_agent_instance_dict = {"agent": None} # To store agent instance across calls if needed, or re-init

            async def handle_eb_agent_chat(user_message: str, chat_history_list: list, current_token_state: dict):
                # Expected outputs: [eb_agent_chatbot_ui, eb_agent_chat_history_st, eb_agent_chat_input_ui, eb_agent_status_md, eb_agent_schema_display_md]
                # (5 components)

                if not EB_AGENT_AVAILABLE or not os.getenv('GEMINI_API_KEY'):
                    no_key_msg = "L'Agente AI non è disponibile. Assicurati che GEMINI_API_KEY sia configurata."
                    chat_history_list.append([user_message, no_key_msg])
                    # Yield updates for all 5 components
                    yield chat_history_list, chat_history_list, gr.update(value=""), gr.update(value=no_key_msg), gr.update(value="Nessuno schema disponibile.")
                    return

                if not user_message.strip():
                    # Yield updates for all 5 components
                    yield chat_history_list, chat_history_list, gr.update(value=""), gr.update(value="Stato Agente: Per favore, inserisci una domanda."), gr.update() # No change to schema display
                    return

                status_update_msg = "Stato Agente: Elaborazione della tua richiesta..."
                # Show user message immediately, update status
                yield chat_history_list + [[user_message, None]], chat_history_list + [[user_message, None]], gr.update(value=""), gr.update(value=status_update_msg), gr.update()

                # Prepare DataFrames for the agent
                df_follower_stats = current_token_state.get("bubble_follower_stats_df", pd.DataFrame())
                df_posts = current_token_state.get("bubble_posts_df", pd.DataFrame()) 
                df_post_stats = current_token_state.get("bubble_post_stats_df", pd.DataFrame()) 
                df_mentions = current_token_state.get("bubble_mentions_df", pd.DataFrame())
                
                dataframes_for_agent = {
                    "follower_stats": df_follower_stats.copy() if not df_follower_stats.empty else pd.DataFrame(columns=['no_data_follower_stats']),
                    "posts": df_posts.copy() if not df_posts.empty else pd.DataFrame(columns=['no_data_posts']),
                    "post_stats": df_post_stats.copy() if not df_post_stats.empty else pd.DataFrame(columns=['no_data_post_stats']),
                    "mentions": df_mentions.copy() if not df_mentions.empty else pd.DataFrame(columns=['no_data_mentions'])
                }
                
                schemas_text_for_display = "Schemi DataFrames inviati all'Agente:\n\n"
                from eb_agent_module import get_all_schemas_representation 
                schemas_text_for_display += get_all_schemas_representation(dataframes_for_agent)
                max_schema_display_len = 1500
                if len(schemas_text_for_display) > max_schema_display_len:
                    schemas_text_for_display = schemas_text_for_display[:max_schema_display_len] + "\n...(schemi troncati per la visualizzazione)"
                
                current_agent = EmployerBrandingAgent(
                    llm_model_name=EB_AGENT_LLM_MODEL,
                    generation_config_dict=EB_AGENT_GEN_CONFIG,
                    safety_settings_list=EB_AGENT_SAFETY_SETTINGS,
                    all_dataframes=dataframes_for_agent,
                    rag_documents_df=eb_agent_default_rag_docs.copy(),
                    embedding_model_name=EB_AGENT_EMBEDDING_MODEL,
                    force_sandbox=True 
                )
                
                agent_internal_history = []
                for user_q, ai_r in chat_history_list: # Use the passed chat_history_list
                    if user_q: agent_internal_history.append({"role": "user", "content": user_q})
                    if ai_r: agent_internal_history.append({"role": "assistant", "content": ai_r})
                current_agent.chat_history = agent_internal_history

                try:
                    logging.info(f"Sending to EB Agent. User: '{user_message}'. DF Keys: {list(dataframes_for_agent.keys())}")
                    ai_response = await current_agent.process_query(user_query=user_message)
                    
                    updated_gradio_history = []
                    temp_hist = current_agent.chat_history 
                    for i in range(0, len(temp_hist), 2):
                        u_msg = temp_hist[i]['content']
                        a_msg = temp_hist[i+1]['content'] if i+1 < len(temp_hist) else "Thinking..." 
                        updated_gradio_history.append([u_msg, a_msg])
                    
                    status_update_msg = "Stato Agente: Risposta ricevuta."
                    # Yield final updates for all 5 components
                    yield updated_gradio_history, updated_gradio_history, gr.update(value=""), gr.update(value=status_update_msg), gr.update(value=schemas_text_for_display)

                except Exception as e:
                    logging.error(f"Error during EB Agent processing: {e}", exc_info=True)
                    error_msg = f"# Errore dell'Agente AI:\n{type(e).__name__}: {str(e)}"
                    
                    # Ensure the current turn in chat_history_list reflects the error
                    # The last item in chat_history_list is [user_message, None] from the first yield
                    current_turn_history = chat_history_list + [[user_message, error_msg]] # This might duplicate user message if not careful
                    
                    # Let's reconstruct history carefully to avoid duplicates
                    final_error_history = chat_history_list # This already has [user_message, None] as last item if first yield happened
                    if final_error_history and final_error_history[-1][0] == user_message and final_error_history[-1][1] is None:
                        final_error_history[-1][1] = error_msg # Update the assistant part of the last entry
                    else: # Should not happen if first yield was correct
                        final_error_history.append([user_message, error_msg])


                    status_update_msg = f"Stato Agente: Errore - {type(e).__name__}"
                    # Yield error updates for all 5 components
                    yield final_error_history, final_error_history, gr.update(value=""), gr.update(value=status_update_msg), gr.update(value=schemas_text_for_display)


            def clear_eb_agent_chat_history():
                initial_msg = "Ciao! Sono il tuo Agente AI per l'Employer Branding. Come posso aiutarti?" if EB_AGENT_AVAILABLE else "Agente AI non disponibile."
                return [[None, initial_msg]], [[None, initial_msg]], "Stato Agente: Chat resettata."

            # Connect UI to Handler for EB Agent
            eb_agent_submit_btn.click(
                fn=handle_eb_agent_chat,
                inputs=[eb_agent_chat_input_ui, eb_agent_chat_history_st, token_state],
                outputs=[eb_agent_chatbot_ui, eb_agent_chat_history_st, eb_agent_chat_input_ui, eb_agent_status_md, eb_agent_schema_display_md],
                api_name="eb_agent_chat_submit"
            )
            eb_agent_chat_input_ui.submit(
                fn=handle_eb_agent_chat,
                inputs=[eb_agent_chat_input_ui, eb_agent_chat_history_st, token_state],
                outputs=[eb_agent_chatbot_ui, eb_agent_chat_history_st, eb_agent_chat_input_ui, eb_agent_status_md, eb_agent_schema_display_md],
                api_name="eb_agent_chat_enter"
            )
            eb_agent_clear_btn.click(
                fn=clear_eb_agent_chat_history,
                inputs=[],
                outputs=[eb_agent_chatbot_ui, eb_agent_chat_history_st, eb_agent_status_md],
                api_name="eb_agent_clear_chat"
            )


    # --- Sync Events (at the end of the app's 'with gr.Blocks()' context) ---
    sync_event_part1 = sync_data_btn.click(fn=sync_all_linkedin_data_orchestrator, inputs=[token_state], outputs=[sync_status_html_output, token_state], show_progress="full")
    sync_event_part2 = sync_event_part1.then(fn=process_and_store_bubble_token, inputs=[url_user_token_display, org_urn_display, token_state], outputs=[status_box, token_state, sync_data_btn], show_progress=False)
    sync_event_part3 = sync_event_part2.then(fn=display_main_dashboard, inputs=[token_state], outputs=[dashboard_display_html], show_progress=False)
    sync_event_final = sync_event_part3.then(
        fn=refresh_all_analytics_ui_elements, # This is synchronous
        inputs=[token_state, date_filter_selector, custom_start_date_picker, custom_end_date_picker, chat_histories_st], 
        outputs=apply_filter_and_sync_outputs_list, 
        show_progress="full"
    )


if __name__ == "__main__":
    if not os.environ.get(LINKEDIN_CLIENT_ID_ENV_VAR): logging.warning(f"ATTENZIONE: '{LINKEDIN_CLIENT_ID_ENV_VAR}' non impostata.")
    if not all(os.environ.get(var) for var in [BUBBLE_APP_NAME_ENV_VAR, BUBBLE_API_KEY_PRIVATE_ENV_VAR, BUBBLE_API_ENDPOINT_ENV_VAR]):
        logging.warning("ATTENZIONE: Variabili Bubble non impostate.")
    
    if not EB_AGENT_AVAILABLE:
        logging.error("L'Agente AI per l'Employer Branding non è disponibile a causa di errori di importazione.")
    elif not os.getenv('GEMINI_API_KEY'):
        logging.warning("ATTENZIONE: GEMINI_API_KEY non è impostata. L'Agente AI per l'Employer Branding potrebbe non funzionare.")

    try: logging.info(f"Matplotlib: {matplotlib.__version__}, Backend: {matplotlib.get_backend()}")
    except ImportError: logging.warning("Matplotlib non trovato.")
    
    app.launch(server_name="0.0.0.0", server_port=7860, debug=True)