🏛️ The Last CEO

One CEO. Twenty Years. Infinite Consequences.

An AI-powered business strategy simulator where you lead a company through the AI revolution to 2035 — every quarterly decision is scored by a live XGBoost prediction engine.

🔗 Live app: the-last-ceo-eight.vercel.app  ·  API docs: the-last-ceo.onrender.com/docs

---

📑 Table of Contents

• Overview
• Key Features
• Tech Stack
• Machine Learning
• Game Economy
• Controls
• System Architecture
• Project Structure
• Getting Started
• API Reference
• Deployment
• The Endings Matrix
• Contributors

---

🧭 Overview

As organizations adopt Artificial Intelligence, business leaders face high-stakes decisions about investment, workforce transformation, automation, and risk. Poor calls lead to financial loss, regulatory trouble, and failed transformations — yet most simulations either lack realistic AI scenarios or never show the long-term impact of strategy.

The Last CEO closes that gap. You play a CEO steering a company through AI transformation, making quarterly boardroom decisions while a real machine-learning model forecasts your outcomes. The experience blends four engines into one cohesive game:

Engine	Role
🧠 Machine Learning	XGBoost models predict revenue impact & productivity from real AI-adoption data
⚙️ Business Rules	Translates predictions into financial, workforce, and risk dynamics
🌪️ Dynamic Events	Injects recessions, regulation, GPU shortages, cyberattacks, and viral hits
📊 Executive Reporting	Per-quarter board verdicts, scenario comparisons, and readiness scores

Objective: Navigate technological disruption and survive to 2035 while maximizing company performance.

---

✨ Key Features

	Feature	Description
🏢	Guided Onboarding	A cinematic flow — Landing → Sign In → Boardroom Interview → Avatar Design → Live Dashboard
🏛️	Conversational Board Meeting	An interview with the Chairman, CFO, CTO, CHRO & CRO defines your company and AI posture
🎭	CEO Avatar Skins	6 purely cosmetic skins (Cyberpunk Exec, AI Researcher, Quant, Stealth Agent, and more)
🕹️	3D Playable Office	Walk a voxel CEO around an isometric office — WASD / arrow keys / on-screen joystick, left-click to pan, with fullscreen & ambient-audio toggles — and step up to stations to commit decisions
🛠️	Adaptive Decision Engine	21 strategic moves across 9 categories; the hand re-rolls each quarter and adapts to your state
📈	Executive Dashboard	Revenue, ROI, budget, AI maturity, automation, workforce & risk in a cyber-HUD console
🤖	Live ML Predictions	XGBoost forecasts revenue impact and productivity gain on every decision
🎲	Dynamic Events	Competitor launches, regulation, talent shortages, cyberattacks, viral hits
📝	Executive Reports	Board decisions, A/B/C scenario comparisons, and risk/readiness assessments — exportable to .docx
🏆	8 Distinct Endings	From Unicorn Exit to Rogue AI Singularity to Crash & Burn
🧪	ML Strategy Simulator	A live "what-if" sandbox — drag strategy levers and the XGBoost model re-forecasts revenue, ROI & risk in real time
🔍	Explainable AI (SHAP)	Every forecast is broken down by factor in dollars — see why the model predicted what it did
💬	AI Advisor	An LLM (Groq · Llama 3.1) turns the model's output into plain-English strategic guidance
🗨️	AI Chatbot	An always-on conversational advisor (Groq · Llama 3.1) — multi-turn Q&A on AI strategy, your decisions, and how to play, available on every dashboard screen
🔐	Sign-In Gateway	A login / register flow gates entry to the simulation; each access event is recorded to a backend registry
📱	Responsive UI	Adapts from desktop to mobile — the sidebar collapses into a tap-to-dismiss drawer and dashboards reflow for small screens
⚠️	Crisis Engine	Dynamic project-cost scaling + a crisis mode that reshapes the decision pool when you near insolvency

---

🛠 Tech Stack

Layer	Technologies
Frontend	React 18 · TypeScript · Vite · Tailwind CSS · Zustand · React Three Fiber · Recharts
Backend	FastAPI · Uvicorn · SQLAlchemy · Pydantic
Machine Learning	XGBoost · scikit-learn · SHAP · pandas · NumPy · joblib
Generative AI	Groq API (Llama 3.1) — the in-game AI Advisor & conversational chatbot
Persistence	SQLite
Deployment	Vercel (frontend) · Render (backend)

---

🧠 Machine Learning

The simulation is driven by two gradient-boosted regression models trained on a corporate AI-adoption dataset.

• Dataset: corporate_ai_adoption_dataset.csv — ~200,000 rows × 13 columns
• Algorithm: XGBoost Regression (scikit-learn pipeline: scaling + one-hot encoding)
• Targets: Revenue Impact · Productivity Gain
• Feature engineering: 8 interaction/ratio features (e.g. investment × maturity, automation × investment)
• Leakage control: post-outcome fields (cost_savings, productivity_gain) are excluded from the revenue feature set

Performance

Model	R²	Notes
Revenue Impact	≈ 0.70	Tuned hyperparameters, 1st–99th percentile outlier removal, engineered features
Productivity Gain	≈ 0.96	Leakage-safe feature set

Model compatibility: The bundled .joblib models are serialized with scikit-learn 1.6.1 (pinned in backend/requirements.txt). Newer scikit-learn versions cannot unpickle them — install the pinned dependencies to load them correctly.

Input normalization: The training data is normalized (AI adoption 0–1, automation 0–1, maturity 0–10) while the UI uses human-readable scales (0–5, 0–100, 0–100). The backend rescales incoming inputs to the training distribution before inference, so the model always receives in-range features.

Explainable AI & Advisor

The product doesn't just predict — it explains and advises:

• SHAP attribution (POST /api/explain) — a TreeExplainer decomposes each revenue forecast into per-feature contributions (in dollars), so the UI shows exactly why the model predicted what it did — a glass box, not a black box.
• AI Advisor (POST /api/advisor) — the model's metrics and SHAP breakdown are passed to an LLM (Groq · Llama 3.1), which returns a plain-English strategic briefing. (Requires a GROQ_API_KEY — see Getting Started.)
• Both power the ML Strategy Simulator: a live what-if sandbox where dragging a lever re-runs the model, the SHAP explanation, and the advice in real time.

---

💰 Game Economy

Each quarter, the XGBoost prediction feeds a deterministic business-rules engine that updates your company's finances:

• Model-driven revenue — quarterly revenue is anchored to the model's predicted AI revenue impact (plus an organic, headcount-based baseline), so smarter AI strategy directly grows the top line.
• 10-year LTV ROI — return on investment is measured against a 10-year Lifetime Value of the predicted AI revenue versus capital invested, rather than a single-year snapshot, to reflect the long horizon of AI bets.
• Industry-specific economics — revenue scaling, cost structure, and growth differ by sector (Technology, Healthcare, Finance, Retail, Manufacturing, Logistics).
• Workforce dynamics — automation drives attrition while hiring decisions add headcount; morale reacts to layoffs and productivity gains.
• Dynamic cost scaling — project costs scale to your company's size, so a multi-million initiative isn't instantly fatal to a small startup (and isn't trivial to an enterprise).
• Crisis Engine — when profit turns negative and cash runs low, a crisis mode reshapes the decision pool: expensive expansions are suppressed and recovery/defensive plays are surfaced. Bankruptcy logic is strict — run out of capital and the run ends.
• Win / lose — survive to 2035 to trigger an ending, or hit insolvency for bankruptcy. Your final budget, ROI, headcount, morale, and sector decide which of the 8 endings you unlock.

---

🎮 Controls

The 3D office is fully playable. Walk the CEO to a station (HR, ML, or Boardroom) to commit each quarter's initiative.

Action	Input
Move	W A S D or arrow keys, or the on-screen joystick
Pan camera	Left-click + drag
Fullscreen	Toggle button (top-right of the office)
Ambient audio	Music toggle button

---

🏗️ System Architecture

graph TD;
    A[Player] --> B[React Frontend]
    B --> C[Zustand State Management]
    C --> D[Axios API Call]
    D --> E[FastAPI Backend]
    E --> F[Input Normalization & Feature Engineering]
    F --> G[Dynamic Event Engine]
    G --> H[XGBoost Models]
    H -->|Revenue Prediction| I[Business Rules Engine]
    H -->|Productivity Prediction| I
    I --> J[Updated Company State]
    J --> K[Survival & Risk Scoring]
    K --> L[Executive Report]
    L --> M[Dashboard Update]
    M --> N[Next Quarter Simulation]

Loading

---

📂 Project Structure

The-Last-CEO/
├── backend/                       # FastAPI application
│   ├── app.py                     # Server, endpoints, ML inference & business rules
│   └── requirements.txt           # Pinned backend dependencies
├── frontend/                      # React console dashboard
│   ├── public/                    # Static assets (city backdrop, board avatars)
│   └── src/
│       ├── components/            # HUD cards, charts, modals, sidebar, chatbot widget, 3D CEOModel
│       ├── data/                  # decisions.ts (decision pool) · skins.ts (cosmetic skins)
│       ├── hooks/                 # Game loop, adaptive decision roll, API
│       ├── lib/                   # Axios client & styling utils
│       ├── pages/                 # Landing · Auth · Home · Engine · Outcome · Database
│       └── store/                 # Zustand state store
├── models/                        # Pre-trained models (revenue & productivity .joblib)
├── scripts/                       # train_models.py · train_revenue_tuned.py
└── corporate_ai_adoption_dataset.csv

---

🚀 Getting Started

Prerequisites

• Node.js 18+
• Python 3.10+
• Git
• macOS only: the OpenMP runtime for XGBoost — brew install libomp

1 · Backend

# From the project root — create an isolated environment
python3 -m venv backend/.venv
source backend/.venv/bin/activate        # Windows: backend\.venv\Scripts\activate

# Install pinned dependencies (scikit-learn 1.6.1 is required for the bundled models)
pip install -r backend/requirements.txt

# Run the API
cd backend
uvicorn app:app --host 0.0.0.0 --port 8000 --reload

API → http://localhost:8000 · interactive docs → http://localhost:8000/docs

AI Advisor & Chatbot (optional): the /advisor and /chat endpoints use Groq (llama-3.1-8b-instant). Provide a key from console.groq.com/keys in backend/.env, which is loaded on startup:

# backend/.env
GROQ_API_KEY=your_key_here

All other features — predictions, SHAP attribution, and the strategy simulator — operate without it.

2 · Frontend

cd frontend
npm install
npm run dev

App → http://localhost:5173

3 · (Optional) Re-train the models

# From the project root
python scripts/train_revenue_tuned.py

Parses the dataset, applies feature engineering, trains fresh XGBoost regressors, and saves them to models/.

---

🔌 API Reference

Base URL: http://localhost:8000/api

Method	Endpoint	Description
POST	/predict	Run a revenue & productivity prediction for a scenario; returns metrics + A/B/C investment scenarios
POST	/explain	SHAP feature attribution for a prediction (per-factor contribution in dollars)
POST	/advisor	LLM (Groq · Llama 3.1) strategic briefing grounded in the model output
POST	/chat	Multi-turn conversational advisor (Groq · Llama 3.1) — forwards recent message history and returns the assistant reply
POST	/auth/log	Record a sign-in / register event to the backend registry (CEO_Registry.csv)
POST	/save_game_history	Persist a completed run's quarterly payloads to the ledger
GET	/predictions	Return the 50 most recent stored predictions

Example — POST /api/predict

{
  "industry": "Technology",
  "country": "United States",
  "year": 2030,
  "ai_adoption_level": 3.5,
  "ai_investment_usd": 5000000,
  "automation_rate": 45,
  "employee_ai_training_hours": 120,
  "ai_maturity_score": 75,
  "deployment_count": 10
}

Returns metrics (revenue impact, productivity gain, ROI, transformation score, risk, readiness, board decision) and scenarios (A: maintain, B: +20%, C: +50%).

---

🚀 Deployment

Service	Platform	URL
Frontend	Vercel	the-last-ceo-eight.vercel.app
Backend API	Render	the-last-ceo.onrender.com

The backend binds to the platform-provided $PORT and resolves its database to an absolute path, enabling zero-configuration deployment to managed hosts. CORS is enabled for cross-origin requests from the frontend.

Backend — Render (Web Service)

Setting	Value
Root Directory	Repository root
Build Command	pip install -r backend/requirements.txt
Start Command	cd backend && uvicorn app:app --host 0.0.0.0 --port $PORT
Environment	GROQ_API_KEY — authenticates the AI Advisor (Groq)

Frontend — Vercel

Setting	Value
Root Directory	frontend
Framework	Vite
Environment	VITE_API_URL = https://the-last-ceo.onrender.com/api

---

🏆 The Endings Matrix

Eight outcomes depending on your leadership style and financials by 2035:

	Ending	Condition
🦄	Unicorn Exit	Survive to 2035 with > $3M budget or > 150% ROI
🔔	IPO Public Listing	Survive with ≥ 30 staff and ≥ $2M budget
💼	Megacorp Acquisition	Exit with > $1.5M budget or > 100% ROI
👑	Bootstrap Legend	Complete the run starting from bootstrapper capital
☕	Sustainable Lifestyle	Survive with < 15 staff and ≤ $1.5M budget
🤖	Rogue AI Singularity	Technology-sector startup with > 200% ROI
🤝	Talent Acquisition	Go bankrupt but keep > 50% ROI or > 80% morale
💥	Crash & Burn	Run out of capital before 2035

---

👥 Contributors

Contributor	Focus
@shreyascode11	Frontend & full-stack — game engine, UI, 3D office, integration
@Harishlal-me	Backend & architecture — FastAPI, XGBoost inference, deployment
@KhannakPGupta	Machine learning & documentation — model tuning, notebook, docs
@stmdsaifimaaz-maker	Frontend UI — cinematic landing page redesign & animations
@its-akshdeep06	Branding & QA — application logo and bug fixes

---

Good luck, CEO. The board is waiting.