dusty-nv/jetson-containers — reverse-engineered prompt

Build me a practical tool for NVIDIA Jetson users that makes it easy to build and run CUDA ready containers for AI, robotics, and machine learning.

I want someone with a Jetson to be able to install it, pick what they need like PyTorch, TensorFlow, JupyterLab, LLM tools, vision models, ROS, or simulation packages, and then have the tool build or launch the right container for their JetPack and L4T version without them having to understand all the details. It should include simple commands for install, build, run, tagging, testing, and package discovery.

Please organize the packages in a clean modular way so new AI packages can be added later. Include clear README docs with examples for common workflows, like running JupyterLab, building a PyTorch container, or starting an LLM container. Add sensible defaults, helpful error messages, and basic tests so people know their Jetson setup is working.