Hi everyone! I'm Yogin Langalia, a 2nd year student pursuing B.tech in Information Technology at DJ Sanghvi College of Engineering and BS in Data Science at IIT Madras. I'm interested in the Jupyter-Native Metaflow project for GSoC 2026. I've been using Metaflow for a few time and the notebook friction is real, having to squeeze everything into one cell with NBRunner feels unnatural compared to how you'd normally prototype in Jupyter. The Kale-style cell tagging approach makes a lot of sense here. I've started experimenting locally with a basic

%%step

cell magic using IPython's magic system and got a rough prototype working that registers tagged cells as Metaflow steps. Nothing production-ready yet, but enough to understand how the pieces fit together, cell metadata, the FlowSpec class, and how

nbconvert

can be used for the notebook-to-flow export. Planning to submit a proposal for this. A couple of quick questions: For the notebook-to-flow converter, should the output be a minimal 1. .py file or should it also preserve the notebook's markdown/comments as docstrings? 2. For the DAG widget :- is there an existing Metaflow utility that returns the step graph structure, or would I need to parse the FlowSpec directly? Happy to share my prototype if that's helpful. Looking forward to contributing!