import os import gradio as gr import pandas as pd from openai import OpenAI from inference import run_task, TASKS, API_BASE_URL, API_KEY # -- App Logic ----------------------------------------------------------------- def launch_ceo_simulation(task_id): if not API_KEY: return "ERROR: HF_TOKEN missing.", None, None, None client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY) # State for UI history = [] budget_data = [] rep_data = [] clients_served = 0 initial_budget = None initial_rep = None def on_step(data): nonlocal clients_served, initial_budget, initial_rep if "done" in data: return step = data["step"] obs = data["obs"] thought = data["thought"] if initial_budget is None: initial_budget = obs.budget if initial_rep is None: initial_rep = obs.reputation # Track clients served if data.get("reward", 0) > 0.1: clients_served += 1 # Update history history.append({ "Quarter": obs.quarter, "Budget": obs.budget, "Reputation": obs.reputation, "Market": obs.market_phase.value, "Thought": thought }) budget_data.append([obs.quarter, obs.budget]) rep_data.append([obs.quarter, obs.reputation]) # Yield updates to Gradio try: # We wrap the generator in a yield loop for Gradio gen_data = {"step": 0, "logs": ""} def ui_callback(d): gen_data["last"] = d # Run the task in a separate way or just yield manually # Since run_task is synchronous but we want live updates, # we refactor run_task to be a generator in a real app, # but here we'll simulate the live feel by yielding at the end or # using a simple queue if needed. # For this version, let's keep it robust. import threading import time results_queue = [] def queue_callback(d): results_queue.append(d) thread = threading.Thread(target=run_task, args=(client, task_id, queue_callback)) thread.start() while thread.is_alive() or results_queue: if results_queue: data = results_queue.pop(0) if "done" in data: yield ( f"### Task Complete! \n**Final Score: {data.get('score', 0):.3f}**", pd.DataFrame(history), pd.DataFrame(budget_data, columns=["Quarter", "Budget"]), pd.DataFrame(rep_data, columns=["Quarter", "Reputation"]) ) break obs = data["obs"] on_step(data) nw_growth = ((obs.budget - initial_budget) / initial_budget) * 100 if initial_budget else 0 rep_growth = ((obs.reputation - initial_rep) / initial_rep) * 100 if initial_rep else 0 status_md = f""" ## 🏢 CEO Command Center | {task_id} **Net Worth Growth**: `{nw_growth:+.1f}%` | **Reputation Growth**: `{rep_growth:+.1f}%` | **Clients Served**: `{clients_served}` --- **Quarter**: {obs.quarter} | **Budget**: ${obs.budget:,.0f} | **Reputation**: {obs.reputation:.2f} **CEO THOUGHTS**: > {data['thought']} """ if data.get("done"): status_md = f"# 🏆 MISSION ACCOMPLISHED!\n## FINAL GROWTH: {nw_growth:+.1f}% \n{status_md}" yield ( status_md, pd.DataFrame(history), pd.DataFrame(budget_data, columns=["Quarter", "Budget"]), pd.DataFrame(rep_data, columns=["Quarter", "Reputation"]) ) time.sleep(0.1) except Exception as e: yield f"Error: {str(e)}", None, None, None # -- UI Layout ----------------------------------------------------------------- with gr.Blocks(title="AI CEO Dashboard") as demo: gr.Markdown("# 🏢 AI Strategic CEO Dashboard") gr.Markdown("### Meta PyTorch Hackathon — Business Simulation") with gr.Row(): task_input = gr.Dropdown(choices=TASKS, value=TASKS[0], label="Select Business Task") run_btn = gr.Button("🚀 Start CEO Strategy", variant="primary") status_display = gr.Markdown("Click Start to begin the simulation...") with gr.Tabs(): with gr.TabItem("📊 Financial Pulse"): with gr.Row(): budget_plot = gr.LinePlot(x="Quarter", y="Budget", title="Budget Growth") rep_plot = gr.LinePlot(x="Quarter", y="Reputation", title="Reputation Trend") with gr.TabItem("📋 Strategy Log"): log_table = gr.Dataframe(headers=["Quarter", "Budget", "Reputation", "Market", "Thought"], wrap=True) run_btn.click( fn=launch_ceo_simulation, inputs=[task_input], outputs=[status_display, log_table, budget_plot, rep_plot] ) if __name__ == "__main__": demo.queue() demo.launch(server_name="0.0.0.0", server_port=7861)