This is an early release. Staticware does something very satisfyingly today: it serves static files with content-hashed URLs for cache busting. That means when you edit your CSS then redeploy and restart your server, visitors get the latest CSS without forcing a refresh. More will come.
BinaryOrNot identifies binary files three ways: by extension, by file signature, and by content analysis. Pass it any file path and it tells you binary or text, accurately, across PNGs, PDFs, executables, archives, fonts, CJK-encoded text, and hundreds of other formats.
If you've ever had BinaryOrNot misidentify a UTF-16 file, choke on a CJK-encoded document, or crash because chardet changed its API, this release is for you.
You know that thing where you release an album, it's on the shelves, people are buying it, and then someone points out the spine says it's your previous album? That's what happened with Cookiecutter 2.7.0. We put out the long-awaited release with 27 improvements and 17 contributors, and cookiecutter -V proudly announced: **2.6.0**.
Cookiecutter 2.7.0 is tested on Python 3.10 through 3.14, ships with a security policy documenting the trust model for template hook scripts, and publishes to PyPI with cryptographic provenance so you can verify every release. Seventeen contributors from the community helped build it.
Air Web Framework exists not just because we created it, but because of the support we got soft-launching it at Python Philippines and running its first sprint at PyCon Davao. Now, Air is a PythonAsia 2026 community partner.
I've put out the biggest release of cookiecutter-pypackage since the modern rewrite. Generated projects now ship with a documentation site, type checking, cross-version coverage enforcement, and security-hardened CI, all configured and working out of the box.
We recently reviewed a pull request from contributor Martin Saizar that refactored our Python tests. The goal was to enable the PyTest (PT) linter rule in Ruff, our code formatter and linter.
On a recent flight from Japan to the Philippines, Pydanny and I started coding a new project. The result is **AirDragon**, a layout and component library for the Air web framework.
I co-led the first ever Air web framework sprint in Davao, Philippines, as part of Day 2 of PyCon Davao on October 26, 2025.
Slides from my keynote at PyCon Davao 2025, the first regional Filipino conference for Python programming enthusiasts.
Air is a pioneering AI-first Python web framework that represents the next evolution in web frameworks to be deeply AI-native.
If you need to run Python applications from PyPI without installing them permanently, uvx provides a convenient way to execute Python tools without having to manage dependencies manually.
How to get JSON into a local SQLite database fast.
Continuing the Sentence Transformers exploration, I use a cross-encoder model to rank my notebooks by similarity to search queries.
Here I use sentence transformers and a bi-encoder model to encode my notebooks as embeddings and implement semantic search.
When working in notebooks, you'll occasionally need to access the notebook's own filename. Here's code to do it, which I've verified works in Jupyter Notebook Classic.
Happy Pi Day. Here's pi from the stdlib math module and from Numpy.
How to remove stale .pyc files and __pycache__ directories, using uvx + pyclean.
Here I create a local titlecase model based on Mistral, using a local Ollama modelfile and a solid prompt.
Title-casing text is one of those hard problems no one ever gets right, yet no one considers worthy enough to solve with AI. Here I experiment to see if I can improve upon the latest best solutions with a local Ollama modelfile and a solid prompt.
I thought I had completely lost my new iteration on my notebook titler tool, but it turns out I'm finding bits and pieces in various places. Here I try to put it together again.
Here I play with using Pynput as a system level keylogger, from within my Jupyter notebook version of this.
Fastcore's call_parse decorator turns Python functions into CLI tools quickly. It's nowhere near as fancy as Typer or Click, but it's super quick to use.
Here I turn this Jupyter notebook into a Python script, using nbdev's nb_export function from the notebook itself.
I use Gemini's text-embedding-004 model to generate embeddings for sentences. Then I define a cosine similarity function and see what it returns for a few related and unrelated sentences.
I'm starting to accumulate many UntitledX.ipynb files. Here I use the Gemini 1.5 Flash language model from Google to rename each one based on its contents.
This is a proof-of-concept of moving MonsterUI's HTML class injection from lxml post-processing to a custom Mistletoe renderer. Looks like we can get 3x faster Markdown rendering using this simple MonsterHTMLRenderer vs. parsing the rendered HTML to inject the classes.
I define various useful datetime utilities with the help of fastcore's L and map, and functools' partial. Then I extend that to generate Discord timestamps, which localize Unix timestamps to the reader's timezone.
Using Pygments, CSS and Ruff to improve how code blocks are displayed on my daily notebook blog.
Here I use the Python Imaging Library to create and display an image in-notebook, so that it's rendered as part of the blog post on audrey.feldroy.com.
Can we customize a FastHTML app to set different headers when rendering notebooks with nb2fasthtml, based on what the notebook actually needs for its headers?
Working with Anki flashcard decks in Python, with genanki to work with the decks and fastcore for ease of use.
SQLite full text search setup via APSW for all the notebooks on this website, inspired by the APSW FTS5 Tour).
The SQLite FTS5 (full-text search) extension includes the built-in tokenizers unicode61, ascii, porter, and trigram, implemented in fts5_tokenize.c. APSW provides a Python API to use those.
In this tutorial we'll look at the simplest routes and route handlers you can create with FastHTML. We'll define the handlers as little functions, and then call them as we would any other Python function. After that, we'll make simple GET requests to a simple index route/handler, a parameterized one, and a parameterized one with a redirect.
I'm starting with a snippet of code from the bottom of execnb's 01_nbio.ipynb, breaking it down into pieces and seeing each part.
FastHTML provides default headers for every page, which are also fully customizable. This notebook explores how this works.
The zip docs say that zip(*iterables, strict=False):
This notebook shows how to:
## Understand the Problem
This notebook is a SolveIt-style exploration of https://github.com/AnswerDotAI/execnb/. Here I am following the SolveIt process in a Jupyter notebook to learn new things.
Having fun with fastcore's L (list) class and holiday lyrics.
Exploring how execnb reads notebooks and nb2fasthtml renders them, to understand the tools I'm building on.
Here we are checking the numbers from our daughter's snowman card to Daddy. She gave him math problems to solve and a snowman joke.
I'm here to learn the new web framework FastHTML by Jeremy Howard, and I'm using Claudette to help me explore it.
Studying https://claudette.answer.ai/
Studying https://docs.fastht.ml/tutorials/by_example.html more
My study of the @delegates decorator and GetAttr from fastcore.
My early attempts to figure out auth in FastHTML. Things have likely changed a lot since.
My early exploration of the xtend notebook in FastHTML.
Reading through fastcore's xml.py source to figure out how FastHTML components print and render, and what to_xml and highlight do.
Caution: I’ve learned better patterns since I wrote this. Leaving this here for posterity.
I feel like Jupyter notebooks would be really nice for blogging or publishing "Today I Learned" posts. I had heard about Fastpages before via Jeremy Howard's blog or YouTube videos, but seeing that it was deprecated in favor of nbdev, I decided to try nbdev.
In data science, EDA is an exploratory analysis of a data set. The goal is to better understand the stories that the data tells, and to uncover interesting ideas that may turn into new hypotheses to explore.
SciPy has tools for creating Voronoi tessellations. Besides the obvious data science applications, you can use them to make pretty art like this:
In Django terms, a QuerySet is an iterable of database records. What's nice about them is that they are evaluated only when you're ready for the results.
A common thing to do in Python is to go through a directory tree, opening each file and doing something with the file's text. Here's what to do when you hit a UnicodeDecodeError from accidentally opening a binary file.
Daniel and I just returned from a trip to San Antonio, Texas, where we taught one of our intensive Django training workshops at Lackland Air Force Base.
This weekend, Daniel and I drove down to Ensenada, Mexico to speak and coach at DjangoGirls Ensenada. It was a 2-day workshop for women of any level of experience to get a taste of web application development.