trevor manz

@manzt.sh

515 Followers 577 Following

middle child, creator of anywidget.dev building @marimo.io; prev @harvard.edu

Code snippet showing a test for int32 columns. Three variables — col, arr, val — each have a twoslash type snapshot comment below them showing the inferred TypeScript type (Column, Int32Array, number). Below that, runtime assertions check that the actual values match: arr is an instance of Int32Array and val is typeof number.
Code snippet showing three levels of schema strictness in quiver. q.js.number() returns a broad union of typed arrays from toArray(). q.int() narrows to integer typed arrays only. q.int32() narrows all the way to Int32Array. All three return number from at().
Code snippet showing quiver's API: import quiver, define a table schema with int32, nullable utf8, and float64 columns, then call parseIPC on bytes. Accessing row values returns fully typed results — name is string | null, score is number.
A marimo notebook cell showing a TinyVGG model rendered as a collapsible tree. The header shows 8.5M params and 32.3 MB. Layers are color-coded: Conv2d and Linear in blue, BatchNorm2d in green, ReLU in orange, Dropout in red. Each layer shows its config and parameter count.
A marimo notebook cell showing a TinyVGG model rendered as a collapsible tree. The header shows 8.5M params and 32.3 MB. Layers are color-coded: Conv2d and Linear in blue, BatchNorm2d in green, ReLU in orange, Dropout in red. Each layer shows its config and parameter count.
A play on Magritte’s painting: Python code df = df[df["a"] == 1] above the caption “Ceci n’est pas une pipe.”
Screen shot of an email with text highlighted. Subject: Zarrita application The last few months I've been participating in ECMWFs code for earth program. In this project I've been building a CAMS dataset visualization heavily based on your work on Zarrita. (highlighted text) I really enjoyed working with Zarrita and would not have been able to complete the project without your great work. Thank you!
introduction marimo notebook within vscode
minimap popup on notebook helps visualize and navigate cells and their dependencies
Adding and removing packages inside the integrated marimo package manager tree viewer.
Screen recording of a Marimo notebook running in the browser. A cell errors due to a missing package with extras. A UI appears prompting the user to select optional extras before installing. The user selects extras and installs the package directly from the notebook.
Terminal window showing `uvx --with=deno python`, invoking the Deno CLI as a Python package within the Python REPL.
sorting row order of an bar plot in marimo notebook using a custom anywidget
a speaker at PyCon demos using a screen reader on data dashboards, showing how many charts are just announced as “image” and not perceivable.
A parody of Drake’s “If You’re Reading This It’s Too Late” album cover. The text is changed to: “If You’re Reading This Use uv.” Handwritten font mimics the original style.
Terminal screenshot showing the output of `npx @manzt/uv --help`, which displays the help menu for uv, a fast Python package manager.
JavaScript source code containing both `import` and `require` statements in addition to hybrid `import`/`require` lines.
VS Code screen capture using the juv extension. Selecting the `juv` item in the notebook toolbar prompts the user to sync, add, or remove dependencies for a Jupyter notebook. Adding the dependency "anywidget" creates PEP 723 inline metadata, which is used to create an environment to run the notebook.
A TypeScript example comparing type assertions vs. runtime assertions. The first approach (value as number) blindly tells the type checker that value is a number, risking runtime errors if it's null. The second approach uses an assert function to enforce the check at runtime, ensuring value is actually a number before proceeding.
A `deno bench` benchmark on an Apple M3 Max (Deno 2.2.3) shows DecompressionStream is 1.48x faster than fflate.gunzip and 2.42x faster than pako.inflate.
Code snippet decoding a gzipped response with: ```js const compressedResponse = await fetch("/data.gz"); const decompressedResponse = new Response( compressedResponse.body.pipeThrough( new DecompressionStream('gzip') ) ); const text = await decompressedResponse.text(); ```
`juv venv` in VS Code, showing the derived virtual environment selected and running a notebook in the Jupyter extension.
cell imaging dataset with labels overlaid on top.
Side by sider browser console of in JupyterLab with a custom widget, with and without "-jlab_core" filter. The window with the filter hides all the warnings from JupyterLab.
anywidget.dev website screenshot with annotation arrow pointing to a a new github link
printing out a lockfile in the terminal for a Jupyter notebook. Terminal command: cat Untitled.ipynb | jq -r '.metadata["uv.lock"]' | bat -l toml
Code snippet showing TypeScript error handling. A try/catch block uses 'except' to filter for specific error types (DatabaseError, NetworkError) while letting others bubble up. The error variable is automatically narrowed to the matched types, similar to Python's except clause.
the ghostty terminal running a custom shader animation to make text look like its underwater
A Jupyter console session loads Seattle weather data into a DataFrame and uses MosaicWidget to display an interactive "Seattle Weather" visualization. The widget appears in a sidecar window, allowing exploration while the console remains active.
A unix pipe in the terminal adding dependencies to a script. The command is: echo 'nbformat\npillow' | uv pip compile --no-deps - | grep '==' | xargs uv add --script foo.py. And the Python file has the exact versions pinned (e.g., nbformat==5.10.4)
Terminal session showing juv stamp commands with a Python script (foo.py). Initializes the script, stamps the current time, sets a date (2006-07-08), stamps with a Git revision (7fdf3f), updates with the latest commit timestamp, and clears the timestamp.
editing the cell contents of Untitled.ipynb in neovim as markdown with `juv edit Untitled.ipynb`.
side by side of `juv cat Untitled.ipynb` and `juv cat Untitled.ipynb --script`, showing markdown and python script previous in the terminal.
Configuration in starship.toml and .zshrc to display the Python version used by uv in Starship's prompt by repurposing pyenv support. The starship.toml section enables pyenv version display with a uv prefix, and the zshrc function overrides pyenv to return the Python version from uv.
Two consecutive runs of `juv run foo.ipynb` - first with default mode showing verbose Jupyter logs, followed by managed mode showing minimal, clean output.