Sign in

dhirschfeld.bsky.social

@dhirschfeld.bsky.social
81 followers 199 following 159 posts
PostsRepliesMedia
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 09/10/2026
Isaac ROS 5.0 is out and you can install it with Pixi! prefix.dev/blog/isaac-... #NVIDIA #IsaacROS #ROS #AI
prefix.dev
Isaac ROS 5.0 with Pixi
Making Isaac ROS installations easy, even on a Jetson device with Pixi
021
Reposted by @dhirschfeld.bsky.social
matty @mattio.bsky.social · 08/10/2026
This is wild - somebody did a deep dive into the real stories behind this famous XKCD: sheets.works/data-viz/hol...
an XKCD comic showing a precariously assembled tower of blocks. much is labelled "all modern digital infrastructure." a thin leg down at the bottom is labelled "A project some random person in Nebraska has been thanklessly maintaining since 2003."
251290618
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 04/10/2026
The LLMs slowed to a crawl as they tried to understand and reason about the codebase, and consequently the costs increased and development velocity slowed as it became and endless cycle of bug fixes. The solution was to nuke the entire ill-concieved feature, but LLMs can't make that judgement call.
000
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 04/10/2026
I have seen this myself in even small features where every attempt to fix a bug resulted in one (or more) new bugs being introduced. It was a downward spiral of adding more code to try and fix a problem and instead adding new bugs.
110
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 04/10/2026
Because the cost of paying off tech-debt is much smaller, it enables you to take more on, but the costs of taking-on tech-debt haven't change, and those costs increase *exponentially* the more you take on. At some point even a frontier model can't keep up.
110
Reposted by @dhirschfeld.bsky.social
Blosc Development Team @blosc.org · 30/09/2026
📢 Python-Blosc2 4.14 is out! A common interface to stream, query & slice remote Parquet, #PyTables, #HDF5, #Zarr & #Blosc2 data over HTTP/S3: ⚡️ Zero full downloads ⚡️ Cloud index pushdown ⚡️ DuckDB/Polars streaming! Read the tour: blosc.org/posts/remote... Compress Better, Stream Smarter 🚀
012
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 01/10/2026
DuckDB. I run DuckDB. 🦆
000
Reposted by @dhirschfeld.bsky.social
DuckDB @duckdb.org · 25/09/2026
Did you know that, since #DuckDB v0.10.3 (~May 2024), you can point a SELECT at a dataset on the Hugging Face Hub⁠, using the DuckDB `hf://` protocol, and query it, without downloading it first? 🤗 🦆 This blog post covers how that Hugging Face + DuckDB integration works and the use cases it fits:
duckdb.org
DuckDB and Hugging Face: Querying Datasets Directly
Hugging Face hosts hundreds of thousands of datasets, and DuckDB can read them directly, exactly where they are and without downloading anything, over the DuckDB hf:// protocol. This post looks at how...
1203
Reposted by @dhirschfeld.bsky.social
David Brochart @davidbrochart.bsky.social · 25/09/2026
AnyIO might be the single most underrated Python library: only 2.5k stars on GitHub but downloaded over a billion times per month on PyPI!
111
Reposted by @dhirschfeld.bsky.social
Michał (Mike) Krassowski @krassowski.bsky.social · 24/09/2026
Checkout the list of 100+ #jupyter #ai extensions I collated at github.com/openteams-ai...
github.com
GitHub - openteams-ai/awesome-jupyter-ai: AI extensions for JupyterLab and Jupyter Notebook 7
AI extensions for JupyterLab and Jupyter Notebook 7 - openteams-ai/awesome-jupyter-ai
044
Reposted by @dhirschfeld.bsky.social
Quansight @quansight.com · 18/09/2026
Small differences in engine semantics create big shifts in query output. Marco Gorelli breaks down the operational differences between Polars and traditional SQL. Read Marco's technical breakdown & migration tips: labs.quansight.org/blog/polars-... #Polars #SQL #DataScience #OpenSource #Narwhals
labs.quansight.org
041
Reposted by @dhirschfeld.bsky.social
Project Jupyter @jupyter.org · 17/09/2026
🤖 Build, export, assemble, and test a robot model right in your browser. Yahia Heni shares a new JupyterCAD robotics workflow using JupyterCAD-URDF + JupyterLab-URDF, with no local ROS installation required. blog.jupyter.org/a-robotics-w... #JupyterCAD #Robotics #OpenSource
042
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 12/09/2026
I think it is the best way to do it - good software engineering starts with good packaging practices. I think probably no one is do the same though, so it's not an established practice/tool. It's more my own DIY solution.
010
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 12/09/2026
It also reuses the same infrastructure as our packaging so there only one tool handling the building, publishing and installing of packages or agents, or skills.
100
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 12/09/2026
Other than being harness-agnostic, it lets you strongly version your skills and specify depenencies such as other skills or binaries your skill depends on.
100
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 12/09/2026
We package our agents and skills as `conda` packages and use a `pixi-skills` CLI to symlink them for your agent harness of choice.
111
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 12/09/2026
I use a custom fork of `pixi-skills` pavel.pink/blog/pixi-sk...
pavel.pink
Managing Agent Skills with Your Package Manager | pavel.pink
AI coding agents use skills — markdown files that teach them domain-specific tasks. We publish ours as conda packages and manage them with pixi, getting versioning, lockfiles, and supply chain securit...
100
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 06/09/2026
Guess I need actually test it out myself, on my own problems...
010
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 06/09/2026
In that it has broadly similar capabilities and the main reason it was release was to address copyright infringement claims.
110
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 06/09/2026
Isn't it just a limited version of Fable 5?
110
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 31/08/2026
Pixi now fully supports archspec, so you can create CPU-optimized environments and let Pixi handle the installation logic. Read more about it in the blogpost we updated: prefix.dev/blog/buildi...
042
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 27/08/2026
Any Rust, C++ or Python project on GitHub can become a global tool on your machine. No recipe, no manifest in the repo. Just name the backend.
081
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 25/08/2026
Textual! 🤦‍♂️ The library is called textual. textual.textualize.io
textual.textualize.io
Textual - Home
Textual is a TUI framework for Python, inspired by modern web development.
000
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 25/08/2026
Pretty amazing the things you can do with textualize!
110
Reposted by @dhirschfeld.bsky.social
Project Jupyter @jupyter.org · 21/08/2026
JupyterLite 0.8 is out: shareable REPL links, a refreshed loading screen, file system fixes, an in-browser terminal with real WebAssembly tools (git, vim, grep), more kernels via jupyterlite-xeus, and Python 3.14 on Pyodide. Try it with no install: jupyter.org/try-jupyter
072
Reposted by @dhirschfeld.bsky.social
Jaime Rodríguez-Guerra @jaimergp.dev · 21/08/2026
CEP 48, which brings conditionals and extras to your @conda.org dependencies, is now accepted! This means you will soon be able to type things like `typing-extensions[when='python<=3.10']`  and `scipy[extras=test]` 🚀 github.com/conda/ceps/b...
github.com
196
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 21/08/2026
Conda packages have extras now. Optional dependency groups you declare and then consume, via v3 repodata. Plus when conditions and variant flags.
0126
Reposted by @dhirschfeld.bsky.social
Quansight @quansight.com · 21/08/2026
Polars and SQL operate on fundamentally different mental models. Marco Gorelli explores the architectural nuances that shape both paradigms. Understanding these mechanics is key to writing clean code that generalizes seamlessly across query engines. Read Marco’s full analysis: buff.ly/J31DcJK
011
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 20/08/2026
[system-requirements] is deprecated. Platforms now carry the requirements themselves. CUDA and glibc constraints inline, and you can name a target what you actually call it: jetson-nano.
052
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 20/08/2026
OSS collaboration is great to see. I know Nathan has been a core part of moving the scientific Python ecosystem forward! :heart:
010
Reposted by @dhirschfeld.bsky.social
Hynek Schlawack @hynek.me · 20/08/2026
Happy/nervous to announce argon2-cffi-bindings, the low-level C binding wheels used by argon2-cffi. It's a big release w/ wheels for new platforms, but the biggest item is the liberation from setuptools w/ great help by Nathan Goldbaum & @henryiii.bsky.social! 🥳 #Python github.com/hynek/argon2...
github.com
Release 26.1.0 · hynek/argon2-cffi-bindings
Highlights Big and scary release! We've switched from setuptools to scikit-build-core and added wheels for pyemscripten and (provisionally) RISC-V. Full changelog below! Special Thanks This release...
141
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 20/08/2026
I use github.com/sharkdp/fd which, other than being much faster than find, it also ignores git ignored files and folders by default.
github.com
GitHub - sharkdp/fd: A simple, fast and user-friendly alternative to 'find'
A simple, fast and user-friendly alternative to 'find' - sharkdp/fd
010
Reposted by @dhirschfeld.bsky.social
Journal of Open Source Software @joss-openjournals.bsky.social · 19/08/2026
Just published in JOSS: 'JupyterGIS: A Collaborative GIS Environment for JupyterLab' doi.org/10.21105/joss.10383
054
Reposted by @dhirschfeld.bsky.social
Sylvain Corlay @sylvaincorlay.bsky.social · 17/08/2026
Ten years ago, I founded QuantStack on a conviction: that we could build a sustainable business by supporting world-class open-source software to advance science. We had no outside funding, no guarantees, just trust in open collaboration. Read the full story: quantstack.net/blogs/Decade...
quantstack.net
QuantStack
A team behind major open-source projects of the scientific computing ecosystem such as Jupyter, Conda-Forge, Mamba, XTensor/SIMD, Apache Arrow/Parquet and more.
1147
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 13/08/2026
Ten years on, I think the concept of "Cognitive Debt" also deserves a mention. I have been convinced that it is actually a separate thing, distinct (but similar to) technical debt.
000
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 13/08/2026
A long time ago, I was planning on documenting my thoughts on technical debt, but @nina.codes beat me to it, said everything I wanted to say (and did a much better job of it than I ever could)! These days I jut point people to her brilliant talk: "Technical Debt - The Code Monster in Your Closet"
nina.codes
Technical Debt — The Code Monster in Your Closet — PyCon US 2015
Technical Debt — The Code Monster in Your Closet — presented at PyCon US 2015
110
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 13/08/2026
Which means, at some point, not even swarms of mythos-class agents will be able to prevent the inexorable, entropic decay of your codebase. Getting agents to "fix" your codebase only delays the inevitable.
000
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 13/08/2026
I have ~forever been on about the dangers of technical debt due to my lived experience of failed codebases. I think understanding tech-debt and it's tradeoffs has never been more important than it is now. Once in a codebase, the cost of technical debt increases exponentially.
200
Reposted by @dhirschfeld.bsky.social
Hynek Schlawack @hynek.me · 13/08/2026
when I said that “even AGI can't solve exponential complexity curves”, I meant something like this: blog.florianherrengt.com/ai-removing-...
blog.florianherrengt.com
AI is removing the middle class of software engineering
AI makes projects with weak engineering culture fail much faster.
1153
Reposted by @dhirschfeld.bsky.social
QuantStack @quantstack.bsky.social · 12/08/2026
Numba now runs entirely in the browser through #JupyterLite and emscripten-forge 🤘 Read Anutosh's post for more details on how we brought Numba's JIT compiler to #WebAssembly, enabling a broader ecosystem including PyTensor, PyMC, Interpolation․py, and Dolo․py. notebook.link/blog/numba-i...
notebook.link
Numba in the Browser: Unlocking a New Scientific Python Stack in JupyterLite | Notebook.link Documentation
A JIT compiler and its scientific Python ecosystem running entirely in the Web browser
0125
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 11/08/2026
The ability to specify all the non-python dependencies your script needs is a game changer!
010
Reposted by @dhirschfeld.bsky.social
Quansight @quansight.com · 11/08/2026
44 seconds down to 1.5 seconds. That is the difference free-threaded Python made once the real bottlenecks got fixed. Read how each one was found and fixed. buff.ly/9Z6f5ux
031
Reposted by @dhirschfeld.bsky.social
Daniel Rothenberg @danielrothenberg.com · 11/08/2026
How good are the frontier AI weather forecast models from Google DeepMind, Microsoft, and NVIDIA? See for yourself on @brightband.com's Operational WeatherBench dashboard - owb.brightband.com! We're happy to run these models operationally and share our evaluation stats with the community!
owb.brightband.com
Operational WeatherBench | AI Weather Forecast Leaderboard
Live skill rankings for AI and physics-based weather forecast models, scored daily with standard metrics and an open toolkit.
0135
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 04/08/2026
Some people have kids, others don't?
000
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 04/08/2026
New community tool: a live dashboard for the conda-forge staged-recipes queue. See which new packages are waiting for review, what's passing CI, and where things are stuck, all in one place. srdb.thath.net/
043
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 03/08/2026
Will have to check it out as I've had occasion to use freezegun in the past.
000
Reposted by @dhirschfeld.bsky.social
francescalted.bsky.social @francescalted.bsky.social · 31/07/2026
Most compression libraries ask you to move in. We think that's backwards. Blosc2 4.9.1: DuckDB, Polars and pandas 3 read a CTable directly via Arrow's PyCapsule protocol — no conversion step. Cooperation, not completeness. blosc.org/posts/not-an... Compress Better, Compute Bigger 🚀 #DataScience
blosc.org
Not an island: bringing compression to the tabular ecosystem
A compression library can go one of two ways: try to be everything (its own dataframe, its own query engine, its own format that nothing else reads), or be a fast, compact layer that slots underneath
022
Reposted by @dhirschfeld.bsky.social
Prefix.dev - the Pixi company @prefix.dev · 28/07/2026
The rattler-build playground got a bit fancier! Now also generates recipes for PyPI, CRAN, and CPAN upstream sources. Try it out under playground.rattler.build
183
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 26/07/2026
I think Terrance Tao can elicit the results he does because every question is loaded with expertise. I observe the same in software engineering where AI can amplify expertise.
000
dhirschfeld.bsky.social @dhirschfeld.bsky.social · 26/07/2026
Fascinating to see how AI is used in mathematics by true experts! 🤯
120