ZenML

About the project

ZenML is an open-source MLOps framework built and maintained by zenml-io, not by me. It models each step of an ML pipeline as a versioned, cached, typed function so the same pipeline runs locally, on Kubernetes, or against a managed orchestrator without changing the step contract.

ZenML ships an official VSCode extension that lets you browse, run, and inspect pipelines without leaving the editor.

What I built

I built a chatbot feature for the extension.

• Built a chatbot panel inside the VSCode extension that takes natural-language questions about a pipeline (why a step failed, where caching is or isn’t kicking in, how to optimize a run) and answers them with pipeline context attached.
• Integrated multiple LLM providers (OpenAI’s ChatGPT and Anthropic’s Claude) behind a single chat interface so users can pick the model.
• Built the chat UI: dropdowns and interactive controls for selecting pipeline context, plus the state management that keeps that context scoped per conversation.
• Implemented real-time message streaming so responses arrive token-by-token rather than as a single blocking call, with the UI updating live.
• Hardened the LLM-call boundary (failed and partial responses, rate limits, provider errors) and surfaced clear feedback in the panel so the failure mode is never a silent hang.