⚡ NVIDIA cuOpt Scheduler

A high-performance, GPU-accelerated staff scheduling application powered by NVIDIA cuOpt and Streamlit.

This project demonstrates how to solve complex combinatorial optimization problems (like nurse rostering) in seconds using the massive parallel processing power of local NVIDIA GPUs (Ampere/Ada Lovelace architectures).

🚀 Features

• GPU Acceleration: Utilizes NVIDIA cuOpt (via PuLP) to solve scheduling problems significantly faster than traditional CPU solvers (CBC/GLPK).

• Realistic Simulation: Generates "fake" but realistic employee profiles (names, skills, availability) using Faker.

• Interactive Visualization:

  • Gantt Charts: Visual timeline of staff assignments.

  • Heatmaps: Instant analysis of shift coverage and gaps.

• Constraint Handling: Respects maximum hours, skill requirements, availability, and "one shift per day" rules.

• Export: Download optimized schedules as CSV files.

🛠️ Prerequisites

• Hardware: NVIDIA GPU (RTX 30-series or 40-series recommended).
• OS: Linux (Tested on Pop!_OS) or WSL2.
• Drivers: NVIDIA Drivers (535+) and CUDA Toolkit 12/13.
• Python: Version 3.10 or higher.

📦 Installation

1. Clone or Create Directory

   mkdir ~/cuopt-scheduler
   cd ~/cuopt-scheduler

2. Set up Virtual Environment

   python3 -m venv .venv
   source .venv/bin/activate

3. Install Dependencies Note: We must point to the NVIDIA PyPI index to get the optimized cuopt binaries.

   Create a requirements.txt with the following content:

   streamlit
   pandas
   pulp>=2.8.0
   Faker
   plotly
   --extra-index-url [https://pypi.nvidia.com](https://pypi.nvidia.com)
   cuopt-server-cu13
   cuopt-sh-client
   

   Then run:

   pip install -r requirements.txt

🖥️ Usage

1. Activate Environment (if not already active)

   source .venv/bin/activate

2. Run the Application

   streamlit run app.py

3. Interact

   • Open your browser to http://localhost:8501.
   • Use the Sidebar to adjust the number of employees and simulation duration.
   • Toggle "Use NVIDIA GPU" to compare performance against the CPU solver.
   • Click Generate & Solve.

📂 Project Structure

cuopt-scheduler/
├── app.py              # Main application logic (UI, Solver, Viz)
├── requirements.txt    # Python dependencies
├── README.md           # Project documentation
└── data/               # Directory for generated CSVs (optional)

🧠 How It Works

1.Data Generation: The app creates a synthetic workforce with randomized availability and skills (RN, LPN, CNA).

2.Modeling: The problem is formulated as a Mixed-Integer Linear Program (MILP) using PuLP.

• Objective: Minimize total assigned shifts (or maximize efficiency).

• Constraints: Staffing demand, max hours, availability, no double-booking.

3.Solving: * If GPU is selected, PuLP delegates the matrix calculations to cuopt-server, utilizing CUDA cores for rapid convergence.

• If CPU is selected, it falls back to the default CBC solver.

4.Visualization: Results are parsed into Pandas DataFrames and rendered using Plotly.

⚠️ Troubleshooting

"pulp.CUOPT not detected": Ensure you have installed the specific cuopt packages from the NVIDIA index URL. Check that your pulp version is >= 2.8.0.

Optimization Infeasible: If the solver returns "Infeasible," try increasing the Number of Employees in the sidebar. The constraints might be too tight (e.g., needing 5 RNs but only generating 3).

📄 License

This project is open for educational and testing purposes.