From zero to AI inference platform: Building a private AI sandbox on Bare Metal

Every platform team is fielding the same request: "we need somewhere safe to experiment with AI." Third-party APIs raise compliance flags, GPUs are scarce, and hand-built inference boxes turn the platform team into a ticket queue."

This workshop builds a private inference stack live on Latitude.sh bare metal: provisioning a CPU-only server, deploying Ollama to run an open-weight model, and layering Open WebUI on top for secure, multi-user access, then codifying it all so the next sandbox provisions itself.

Attendees leave with a reproducible blueprint, a CPU-vs-GPU decision framework, and the patterns to offer AI sandboxes as a self-service platform product.