Jupyter in the Emacs Universe (martibosch.github.io)

A new blog post surveys how to get a Jupyter-like workflow inside Emacs, why that matters for data scientists, and which Emacs packages work best. It frames Jupyter as three parts — UI, kernel, and the JSON .ipynb format — and reminds readers of a major pain point: many notebooks are irreproducible (a 2019 corpus found only ~3.2% executed to the same results). The author lists desired features for an Emacs setup: mixing code/markdown/plots, proper version control with optional embedded outputs, and IDE-style, environment-aware, notebook‑wide assistance. Configurations and an example notebook/environment are published on GitHub (martibosch/snakemacs and jupyter-emacs-post). Technically, the author compares two main approaches. EIN (Emacs IPython Notebook) emulates the web notebook UI inside Emacs with cell manipulation and inline output, but suffers from limited undo, no Emacs autosave/recovery, and relies on the unmaintained elpy (which can be CPU‑heavy). The alternative combines emacs-jupyter (kernel interaction) with code-cells and Jupytext to convert .ipynb into plain-text .py/.md/.org representations—enabling standard VCS, LSPs like pyright for fast, environment-aware completions, and reproducible scripts while preserving the literate narrative. The post recommends using conda/mamba + conda.el + ipykernel to manage per-buffer environments and suggests LSP integration over elpy for scalable IDE features, making Emacs a viable option for serious AI/ML notebook workflows.