alexwortega/claude-ml-intern-skill

This is a "skill", a set of instructions, that turns the Claude Code AI assistant into an autonomous machine-learning intern. You give it a single instruction like "implement this AI architecture and train it," and it handles the entire process with no further human input: planning the work, researching the model design, writing the code, running tests, training the model, verifying the results, and finally publishing the finished model to Hugging Face (a popular public hosting platform for AI models). The self-verification step is particularly notable: rather than just checking whether the loss number (a measure of training accuracy) looks good, the skill runs six independent checks to confirm the model is actually generating sensible output. This was added after a real failure where good-looking numbers masked a broken model. You install it with a single terminal command, and it slots into Claude Code as a drop-in skill, no separate AI client or extra logins required. Once installed, it triggers automatically when you describe an ML task. At each stage of the process (planning done, code ready, training started, training complete, published), it can send notifications to Telegram or Slack if you configure those. The skill is primarily useful for ML researchers or developers who want to quickly prototype and test new model architectures without spending a day on manual setup. The README notes the full pipeline ran end-to-end on a single GPU in about six hours for a complete training run. The full README is longer than what was provided.