Deploy and Host TorchCode on Railway

About Hosting

TorchCode is a self-hosted, JupyterLab-based coding practice platform focused on PyTorch interview questions. It includes 40 curated notebook problems, built-in judging (torch_judge), hints, and progress tracking.

TorchCode runs as a single container using the upstream image from GHCR. The app serves JupyterLab over HTTP and stores progress in a JSON file. For Railway hosting, you only need one app service with a persistent volume mounted to /app/data.

Why Deploy

• PyTorch interview practice at your own pace — solve common operator and architecture coding questions in a structured environment.
• Self-hosted and private — your notebooks and progress stay in your own Railway project.
• Single-service deployment — no database or cache required.
• Instant feedback loop — run checks, get hints, and track solve history directly in notebooks.
• Upstream image support — deploy from official ghcr.io/duoan/torchcode image with minimal config.

Common Use Cases

• ML interview prep — practice implementing softmax, attention, normalization, and Transformer components from scratch.
• Team study environment — spin up an internal shared TorchCode instance for interview training.
• Hands-on PyTorch learning — use guided notebooks with correctness checks instead of passive reading.
• Reusable coding drills — reset notebook templates and repeat timed attempts.

Dependencies for Deployment

Deployment Dependencies