The data community is coming together in London — and you don’t want to miss it. PyData London 2026 runs June 5–7 at Convene Sancroft, St. Paul’s, and whether you’re a seasoned data engineer or just finding your footing in the world of open-source tooling, this conference has something for you. Here are five reasons to grab your ticket now.

1. Learn from people who are building the tools you use

This year’s keynote lineup is stacked. Samuel Colvin, the creator of Pydantic — downloaded over 280 million times per month and a dependency of virtually every major GenAI Python library, including the OpenAI SDK, the Anthropic SDK, and LangChain — will take the stage. So will Jeremiah Lowin, founder and CEO of Prefect and author of FastMCP, the framework that’s quickly becoming the standard for working with the Model Context Protocol. These aren’t just names — these are the people shaping the infrastructure of modern data and AI work. You’ll hear from them directly, ask questions, and maybe even grab coffee with them between sessions.

2. The agentic web is here — and someone’s going to explain it

Rachel-Lee Nabors has spent their career at the intersection of web standards and developer education — on the React Team, at W3C, and across FAANG companies and startups. Now they’re focused on the emerging agentic web: a future where AI agents browse, build, and interact on the web the way humans do. Their keynote will be a forward-looking session on the future of the open source ecosystem. If you want to understand where this all goes next, don’t miss it.

3. Real-world AI from research to practice — not just theory

Martin O’Reilly, Director of Research Engineering at the Alan Turing Institute, leads the team bridging the gap between cutting-edge AI research and actual deployment — from AI-assisted air traffic control to weather prediction. His work is a reminder that the most impactful data science isn’t happening in a notebook in isolation; it’s happening in teams, institutions, and systems that affect real lives. His keynote will ground the week in what’s actually being built and shipped at scale.

4. Three days of hands-on learning, community connection, and discovery

Friday kicks off with full-day tutorials — deep, hands-on sessions where you actually learn by doing. Saturday and Sunday bring a full program of talks, lightning talks, and live keynotes. There’s space to learn something new, space to go deep on something you’re already working on, and — most importantly — space to meet the people who are working on the same problems you are.

5. London in June. St. Paul’s Cathedral steps away. Enough said.

The venue is Convene Sancroft, right in the heart of the City of London, steps from St. Paul’s Cathedral. If you’re going to spend three days learning, connecting, and thinking about the future of data, there aren’t many places better to do it. Whether you’re local or traveling in, this is the kind of event worth making the trip for.

Tickets are on sale now. Grab yours at the link below, and we’ll see you in June.

→ Get your tickets: ti.to/pydata/pydatalondon26

Spring is here, and the NumFOCUS community hasn’t slowed down one bit. From a feature-packed IDE release to a major data compression milestone, supply chain security upgrades, and a brand-new affiliated project, April has been a busy month across the ecosystem. Here’s a full roundup of everything happening this month.

Sponsored Projects

SunPy & Astropy: Processing Artemis II Solar Eclipse Photos

In a delightful intersection of space exploration and open science, SunPy published a blog post this month walking through how to process the stunning solar eclipse photos taken by the Artemis II crew — using SunPy and Astropy. It’s a vivid demonstration of what these two libraries can do in the wild, with real data from an active NASA mission.

Read the Blog Post →

NumPy: Supply Chain Security and Full Type Coverage

The NumPy team has been busy on two significant fronts this month. First, they’ve completed a major rework of their release process to deliver meaningful improvements for supply chain security — a critical investment for a library that underpins nearly all of scientific Python.

Second, the 2025 NumPy Fellowship has concluded with an impressive outcome: NumPy is now fully type-checked. This milestone brings improved developer experience and editor tooling support for one of the most widely used libraries in the ecosystem.

Release Process →

Fellowship Retrospective →

Spyder 6.1.4: A Feature-Rich Release for Scientific Developers

The Spyder team has shipped version 6.1.4 — just eight weeks after 6.1.3 — bringing a notable set of new features, deep fixes to docstring generation, and expanded support for remote workflows. This release is available now for Windows, GNU/Linux, and macOS.

New features include:

• An option to disable Enter for accepting code completions in the Editor (Preferences > Completion and linting > General)
• SSH config file support for creating connections in Tools > Manage remote connections
• Ability to delete, upload, and download multiple files when working with remote filesystems in Files
• A new Files button to jump to the directory of the currently open Editor file

On the fixes side, docstring generation received a major overhaul:

• Existing docstring sections are now parsed and incorporated
• Return types can be generated from the function body for Sphinxdoc
• Dozens of bugs and formatting issues have been resolved
• The default shortcut was updated to Ctrl/Cmd+Alt+Shift+D to avoid a macOS conflict

Additional notable fixes: macOS standalone app now has microphone and camera access; pyarrow is bundled in standalone installers for Pandas 3.0+ dataframe support; and several remote connection issues have been resolved, including SSH stop errors and JupyterHub connectivity.

Full Changelog →

ITK 6.0 Beta 2: Modern CMake Targets and C++ Modernization

The ITK team is excited to announce the second beta release of ITK 6.0. The headline feature is modern CMake module targets — a significant build system upgrade that makes it substantially easier to link ITK into your projects. Beta 2 also delivers continued C++ modernization, Python enhancements, performance improvements, and an expanded embrace of agentic engineering practices.

As a beta release, the team actively encourages the community to test and provide feedback ahead of the first Release Candidate. Your input now shapes the final release.

C-Blosc2 3.0.0 RC2: Scalable, Robust, and Ready to Ship

The Blosc team has released C-Blosc2 3.0.0 RC2 — the fast, compressed, and persistent binary data store library for C — bringing an important round of polish before the final 3.0.0 release.

RC2 builds on the major features introduced in RC1, including variable-length chunks & blocks (VL-blocks) and improved dictionary compression. This release candidate makes Blosc2 more scalable, more robust, and easier to integrate in modern build environments.

Release Notes →

rOpenSci: April Community Roundup

rOpenSci has published their monthly community news roundup for April. As always, it covers the latest activity in and around the rOpenSci community — from package updates to community discussions. Head to the rOpenSci blog for the full breakdown of what’s been happening across the R scientific computing community this month.

Read the rOpenSci blog →

Julia 1.13.0-rc1: First Release Candidate for v1.13

Julia version 1.13.0-rc1 — the first release candidate for v1.13.0 — is now available. Binaries are available via JuliaUp and for manual download on macOS (Intel and M-series), Windows (x86 and x86–64), glibc Linux (x86, x86–64, AArch64), and FreeBSD (x86–64).

As a release candidate, 1.13.0-rc1 is not considered production-ready. It’s intended to give users — especially package developers — a chance to try out their code with 1.13.0 ahead of the full release. Check the Julia NEWS.md for a full list of what’s new in 1.13.

Note: on Cirrus CI (before that service shuts down on June 1), Travis, and AppVeyor, the 1.13 alias now points to rc1. On GitHub Actions, use ~1.13.0–0 or similar.

Release notes →

Affiliated Projects

Heat 1.8.0: New Release, New LinkedIn, New NumFOCUS Affiliation

Heat, the distributed and GPU-accelerated array library, has had a big April. The team shipped Heat 1.8.0, presented at PyConDE & PyData 2026, launched a new LinkedIn page, and — most excitingly — officially became a NumFOCUS Affiliated project.

Heat 1.8.0 is available now — see the full release notes on GitHub

The team presented at PyConDE & PyData 2026 this week

Follow along on their new LinkedIn page

Welcome to the NumFOCUS family!

Release Notes →

The Sovereign Tech Agency has entered into a service agreement with mlpack to support sustained investment in open-source machine learning infrastructure. The agreement reflects a broader concern raised by the Agency: “the general lack of investment into open-source ML/AI has allowed a few dominant players to play an outsized role in shaping our open digital infrastructure.” This work aims to broaden the ecosystem of tools available to researchers and developers, strengthening open alternatives and ensuring that critical machine learning infrastructure remains transparent, accessible, and community-driven.

What They’re Working On

The service agreement will fund 18 months of focused development on mlpack and its companion library, Bandicoot (a supporting library for high-performance numerical computation).

Over this period, the project will deliver 10 concrete milestones organized around three major goals:

• Expanding hardware support
• Improving interoperability with other tools and ecosystems
• Strengthening mlpack’s documentation for users who rely on it most

The work will be carried out by some of mlpack’s core maintainers, plus contributions from other members of the community. The core maintainers include:

• Ryan Curtin — primary maintainer since mlpack’s founding in 2008; PhD in machine learning
• Marcus Edel — core maintainer since 2013; PhD in machine learning
• Dirk Eddelbuettel — core maintainer of mlpack and Armadillo since 2013; 30+ years of open source experience; Professor of Statistics

The Bigger Picture

This effort has implications across the scientific research community. It reflects a broader push to invest in open, accessible AI infrastructure that researchers and practitioners can rely on.

mlpack is an example of what that infrastructure can look like: an open governance model where maintainers collaborate on a shared roadmap that reflects user needs. NumFOCUS provides institutional support and fiscal sponsorship, ensuring the project has the administrative and financial infrastructure it needs while preserving its independence.

The Sovereign Tech Agency’s investment recognizes that this kind of infrastructure (independent, community-governed, and genuinely open) is worth sustaining. Just as societies invest in roads, protocols, and standards bodies, this work supports the foundational layers of modern computing. Without it, the default trajectory is an ecosystem shaped primarily by a small number of dominant platforms.

The 18-Month Roadmap

The work plan includes 10 milestones across three areas.

On the hardware side, Bandicoot will gain Vulkan, Metal, and HIP/ROCm backends — replacing the deprecated OpenCL standard with modern GPU support across NVIDIA, AMD, and Apple hardware. It will also add low-precision format support (BF16, FP8, FP6, FP4), enabling larger models to run efficiently on smaller devices.

On the interoperability side, the team will develop a production-ready ONNX importer so models trained in PyTorch or TensorFlow can be brought into mlpack more seamlessly. Language bindings for Python, R, and Julia will be refactored into a more consistent object-oriented API. Additional integrations include DuckDB for running mlpack algorithms as SQL table functions and Apache Arrow support for modern data engineering workflows.

On the documentation side, improvements will include enhanced dataset and media loading support, along with a full set of hardware-specific tutorials — from Raspberry Pi object detection to microcontroller inference to GPU fine-tuning on NVIDIA and AMD systems — aimed at making deployment more accessible, especially for embedded and applied ML users.

About mlpack

mlpack was released in 2008 and has been continuously developed under an open governance model. It has been used in research and applied systems, including NASA-related satellite work and low-resource healthcare applications, and has been cited in hundreds of academic publications. Its linear algebra dependency, Armadillo, has been downloaded tens of millions of times.

The project remains intentionally lightweight, prioritizing efficiency and accessibility while maintaining strong performance across constrained environments.

In short, mlpack is well-suited for low-resource and edge AI deployments.

What’s Next

For current users, the next 18 months will bring meaningful improvements to the tools they rely on. For those considering mlpack, the interoperability and documentation work is designed to lower the barrier to entry. And for contributors, the project remains open to participation across code, documentation, and community discussion.

A small group of dedicated open-source maintainers has been building efficient, high-quality machine learning tools for years. This work continues that effort — strengthening the foundations so more people can build on them.

A New Kind of Town Hall

Open source software is everywhere. It powers scientific discovery, financial systems, climate models, and the AI tools reshaping every industry. But ask most people who built those tools whether they can tell you how many people depend on their work, and you’ll often get a shrug.

That’s what the best of open source software (b.o.s.s.) town hall series wants to address. Launched by NumFOCUS in early 2026, b.o.s.s. is a new quarterly education series led by partners and community members. It gives NumFOCUS Sponsored and Affiliated Projects a high-visibility platform to showcase what they’ve built and learn from their community.

The challenge isn’t technical. It’s structural. As James Bednar, Senior Director of Professional Services at Anaconda, put it during the Q&A session, commercial software companies have customers and direct relationships with those customers. Open source tools are just picked up — through word of mouth, through imperfect discovery, sometimes for the wrong reasons. There’s no feedback loop, no direct channel, no easy way to tell your story to the people who most need to hear it.

“Our goal with b.o.s.s. programming is to offer our community a high-visibility platform to learn how to communicate value, impact, and find collaborators for your projects.”
 — MK (Mariel Kanene), Project Resource Mobilization Lead, NumFOCUS

The theme of the first session was centered on marketing as a soft power for open source scientific computing projects — the intentional, strategic work of communicating impact, finding collaborators, and building the kind of visibility that sustains a project over time.

Marketing Starts With Strategy

Marc Ostertag, NumFOCUS Senior Director of Development, put it plainly: marketing requires both a strategy and the right tools to execute it. Most projects, he observed, reach for the tools first and miss the strategy.

“Marketing is, above all, a strategy — one that needs to be carefully thought through and evaluated before turning to the tools that will bring visibility, energy, and momentum to your campaign.”
 — Marc Ostertag, Senior Director of Development, NumFOCUS

The fact is that projects grow when they master both strategy and the right tools, and right now, some projects may struggle to juggle that with everything else they need to do to keep their work moving forward, leading to things getting lost in an oversaturated market.

What NumFOCUS Offers Projects Today

In this market, how do you get started with marketing your OSS project? Kelby Lorenz, NumFOCUS Digital Marketing Specialist, walked through the existing marketing infrastructure for Sponsored and Affiliated projects.

Monthly project updates: Every month, NumFOCUS compiles project news — bug fixes, releases, announcements, anything worth sharing — into a blog post, social media posts for each project, and an email newsletter. The reach: over 800 blog subscribers, nearly 40,000 followers across LinkedIn, X, and Bluesky, and approximately 27,000 email subscribers.

“It’s a really great opportunity to share what you’re doing in a very low-effort way. You just send over the notes, and we take care of the rest.”
 — Kelby Lorenz, Digital Marketing Specialist, NumFOCUS

Social media management resources: NumFOCUS has a detailed slide deck available to all of their Sponsored and Affiliated projects covering target audience research, content calendar setup, analytics interpretation, and visual creation.

One-on-one support from the NumFOCUS digital team: Whether projects are starting from scratch on social media, trying to understand which platforms make sense for their audience, or struggling to interpret what their analytics are telling them, the NumFOCUS team is available to help — from platform selection and content strategy to sitting down and walking through the data together.

Coming Soon: Submit Your Project to the NumFOCUS YouTube Channel

NumFOCUS operates two YouTube channels — one dedicated to PyData conference content, and a second NumFOCUS channel — with a combined subscriber base of over 170,000 and approximately 10,000 views per week..

Projects can now submit short demos and visually engaging content for a regular feature on the NumFOCUS YouTube channel. More details on the submission process and requirements will be available soon. For NumFOCUS projects looking to get started promoting their work, this is an easy way to reach a large audience.

You Can’t Tell a Story You Can’t Measure

Anaconda.org hosts open source packages that power enormous amounts of scientific and data work, but right now, maintainers have limited visibility into that usage. Download counts are flat numbers. There’s little insight into downstream dependencies, geographic reach, or adoption patterns across versions. Without that data, making a case for continued investment — to funders, to employers, to potential contributors — is needlessly hard.

Daina Bouquin, Senior Developer Relations Engineer at Anaconda, has spent her career working on exactly this problem. Before Anaconda, she worked in academia with astronomers and astrophysicists, and she routinely collaborated with groups like Force11 to develop machine-actionable citation principles that enable software contributions to be tracked and credited the same way journal articles are.

“If you can’t answer how many people actually depend on what you built, you’re going to have a hard time making a case for yourself.”
 — Daina Bouquin, Senior Developer Relations Engineer, Anaconda

What Anaconda Is Building with Community Input

Daina walked through several initiatives currently in development — all still being shaped with community input, which she emphasized is deliberate.

Dedicated analytics dashboard: A package and channel maintainer view offering real-time monitoring of what’s happening with a package — architecture breakdown, Python version mix, top-dependent packages, and more. This is explicitly intended to be co-built with the community, and its shape will be informed by what maintainers actually find actionable.

Trusted publishing via OIDC: This top-requested roadmap item will improve publishing pathways for maintainers. Trusted publishing makes it easier to publish packages while maintaining security.

Open API for Anaconda.org: This would make package data accessible in a much more flexible way, allowing maintainers to build custom dashboards, integrate stats into READMEs, and power their own reporting.

AI integrations: As more developers begin their research with AI agents and LLMs, Anaconda is exploring how to surface packages when developers need them, making discoverability part of the infrastructure.

“This data is meaningfully yours. We want to make sure it’s useful so that you can take your package data, build dashboards, generate shareable content, and integrate stats with your README and other reporting requirements.”
 — Daina Bouquin, Anaconda

Anaconda’s commitment: if something they build doesn’t work for you, they want to know. Developer and maintainer feedback is sent directly to the product team.

Lightning Talks from the NumFOCUS Community

The best proof that this work matters is seeing it in practice. Two NumFOCUS-supported projects presented at the inaugural b.o.s.s. town hall — one from Jiangsu, China, and one from Valencia, Spain — demonstrating the community’s global reach and the range of problems it tackles.

SciML

Erik, a researcher within the SciML (Scientific Machine Learning) ecosystem, presented work funded in part by NumFOCUS. His focus: GPU-accelerated boundary value problem (BVP) solvers — a technically demanding but high-impact extension of the SciML project’s existing GPU-accelerated ODE and SDE solver capabilities.

“With the support of NumFOCUS, we can not only complete key technical milestones, but also ensure that these advancements are accessible, well-documented, and sustainable within the open source ecosystem.”
 — Erik, SciML / Samuel Organization

The downstream impact of this work is wide-ranging. Improved BVP solvers open the door to large-scale parameter estimation, symbolic-numerical code generation through ModelingToolkit.jl, and tighter integration with modeling ecosystems like OpenModelica and domain-specific toolchains in the engineering and sciences. The work also advances differentiable programming and optimal control workflows that bridge traditional differential equation solving with modern optimization pipelines.

Support from NumFOCUS makes it possible to take this work from a research prototype to a production-ready tool that’s truly usable and maintainable by the broader community.

Blosc + Caterva2 + LLMS

Francesc Alted, lead developer of Blosc and CEO of Iron Array, presented alongside his colleague Pau to demo Caterva2 — an open-source project built on top of Blosc2 that tackles a fundamental physics problem in data science: at the scale of hundreds of terabytes, you simply cannot download your data to a local machine to process it. The data has to stay where it is, and the compute has to come to it.

Caterva2 makes this practical. It’s an open-source server that supports Blosc2 and HDF5 formats for large dataset storage, with computation, visualization, and filtering capabilities exposed through a REST API, a Python API, and a web interface. The web interface supports running Jupyter notebooks against remotely stored data fully in-browser, with no installation required.

“The user doesn’t need to install anything. Everything is available right in the browser.”
 — Francesc Alted, Lead Developer, Blosc / CEO, Iron Array

The headline feature of the demo was an LLM-powered agent that lets users query datasets using plain natural language. Rather than writing code, a user can ask the agent things like ‘give me the statistical information about this dataset’ or ‘give me a 10×10×10 slice of this array,’ and the agent automatically selects and chains the right underlying Caterva2 tools. Francesc and Pau also walked through more advanced options, including exposing the agent’s internal reasoning so researchers can see exactly which operations are being performed.

On the roadmap: expanded natural language querying across tables and visualizations, prebuilt LLM skills for common scientific computational tasks, and — notably — a Model Context Protocol (MCP) server designed to feed optimal compression parameters across different LLM services. That last item threads directly into the broader ecosystem trend Anaconda raised in the previous section: making open source tools findable and usable at the point where AI agents are doing the work.

Lessons from the Q&A

The sessions above covered strategy, infrastructure, and cutting-edge demos — but the Q&A brought it back to a practical question: once you know you need to tell your story better, what do you actually do?

Daina pointed to the AstroPy and SunPy communities as a concrete example. Rather than hoping users would acknowledge the software in papers, they specified exactly what that acknowledgment should look like: ‘AstroPy Collaboration,’ placed in a particular section of a paper, in a particular format. They made it easy, specific, and trackable so tools like Google Scholar could automatically detect and count those citations.

The lesson for open source projects outside academia: if you want to be found and credited, don’t just ask people to acknowledge you. Give them the exact language and format, and make it as easy as possible to do it right.

“It wasn’t just storytelling to say, ‘Look how impactful we are.’ They went and said: if you want to credit the AstroPy project, say ‘AstroPy Collaboration’ — and this is where we want you to put it in your paper.”

— Daina Bouquin, Anaconda

A practical starting point: CITATION.cff files.

These machine-actionable citation files live in a project’s repository and specify exactly how the software should be cited — in a format that Google Scholar, Zenodo, and others can automatically read and process. Low effort to add; meaningful increase in the likelihood your work gets captured and attributed.

The equity dimension is worth naming directly: the ability to self-advocate isn’t equally distributed. Projects with strong leadership or institutional backing are better positioned to do this work. Many aren’t. That’s part of what makes infrastructure investments like Anaconda’s analytics initiative meaningful beyond convenience — when usage data is surfaced automatically, you don’t have to already know how to advocate for yourself to start telling a data-driven story.

Getting Involved

The gap between the impact open source projects have and the visibility they receive is not inevitable. It’s a solvable problem, and it’s being actively worked on.

NumFOCUS is building the marketing infrastructure and platform access to help projects amplify their work. Anaconda is building the data infrastructure to give maintainers the metrics to back up their stories. The projects presenting at b.o.s.s. are doing the hard technical work and, increasingly, finding the language to talk about why it matters.

What all of this requires from projects themselves is the willingness to invest in the strategy side — to think about who their audience is, what story they’re telling, and how to make it easy for the people who depend on their work to say so. The tools are there. The audience is there. The support is there.

b.o.s.s. will continue quarterly, with new partners, new project demos, and new themes. If you’re a NumFOCUS project and want to participate as a presenter, a future spotlight, or simply as an engaged community member, we’d love to hear from you.

What’s Coming Next

• Daina Bouquin’s Give Me 5 episode (Season 2, Episode 2) recently went live on the PyData YouTube channel, so be sure to give it a watch!

• The NumFOCUS YouTube channel will soon be open for project demo submissions.

• The next b.o.s.s. town hall: June 25, 2026, centered around the theme of A.I. in open source software

The NumFOCUS community has been busy! From milestone releases to Google Summer of Code acceptances, our sponsored and affiliated projects are firing on all cylinders to start 2026. Here’s a roundup of everything happening across the ecosystem this month.

Sponsored Projects

CuPy Hits a Major Milestone with v14

Ten years. 60 million downloads. 10,000 GitHub stars. And now, CuPy v14.

The CuPy team dropped their first major release in two years, and it’s packed. Full alignment with NumPy v2’s updated semantics and type promotion rules brings CuPy in lockstep with the broader Python scientific computing world. The release also introduces initial support for bfloat16 and structured dtypes — exciting steps for users working at the cutting edge of machine learning hardware.

One of the most developer-friendly additions: CUDA pip wheels. You can now install CuPy and the necessary CUDA Toolkit components directly via pip, no system-wide CUDA installation required. For AMD users, a new binary package (cupy-rocm-7-0) adds support for ROCm 7.0 environments. And with over 50 new APIs added — including cupy.linalg.eig and SciPy interpolation routines — there's something here for nearly every use case.

The team also welcomed two new maintainers, Leo Fang and Sebastian Berg, whose expertise across the Python and CUDA ecosystems will help carry CuPy confidently into its second decade. Congrats to the whole CuPy team on a landmark release!

Read the full announcement: Announcing CuPy v14

Python-Blosc2 4.0.0: Blazing Fast with miniexpr

The Blosc development team has released Python-Blosc2 4.0.0, headlined by a powerful new feature: miniexpr, a hyperfast, fully multithreaded computation engine. By computing expressions in blocks of data that fit within private CPU caches, miniexpr can deliver dramatic speedups for the right workloads. The team has also added a new plugin for the OpenZL library, extended blosc2.open() to support .b2z, .b2d, and .b2e files, and introduced None indexing for lazyudf/lazyarray.

Accompanying this release is C-Blosc2 v2.23.0, with the underlying changes needed to power everything in the Python layer.

Learn more about miniexpr: Blosc blog

MDAnalysis Joins GSoC 2026 as a Mentoring Organization

Great news for open-source contributors and mentors alike: MDAnalysis has been accepted as a mentoring organization for Google Summer of Code 2026! If you’re a developer eager to contribute to molecular dynamics analysis tooling — or an experienced community member interested in mentoring the next generation of scientific software developers — this is your moment. Head over to the MDAnalysis Discord or GitHub Discussions to get involved, and check out their blog post for project ideas and details on the contributor application process.

Affiliated Projects

pvlib Rings in 2026 with Two Releases

pvlib came out swinging in early 2026, shipping not one but two new releases before February was barely underway.

v0.14.0 (January 17) brought contributions from 15 community members, adding iotools functions for ERA5 and MERRA2 data, support for estimating single-diode model parameters from datasheet information, accelerated I-V curve calculations, and improved overview documentation for single-diode and temperature models.

v0.15.0 (February 3) followed up quickly to catch a few missed deprecations and update gallery examples for compatibility with pandas 3.

Full details are in the pvlib documentation, and releases are available on PyPI and conda-forge.

Skforecast 0.20.0: 10x Faster Prediction Intervals

The skforecast team has released v0.20.0, with a focus on supercharging probabilistic forecasting. The headline: a 10x speedup in generating prediction intervals, thanks to a full refactor of the bootstrapped residuals calculation across all recursive forecasters. Quantifying uncertainty at scale just got a whole lot more tractable.

This release also introduces native AutoARIMA and AutoETS implementations — highly optimized, scikit-learn-compatible models that drop cleanly into existing skforecast pipelines. And ForecasterStats has been upgraded to support multiple estimators (Sarimax, Arima, Arar, Ets) simultaneously, making benchmarking and model selection a single-call affair.

Check out the full release notes and documentation.

Awkward Array 2.9.0: Stability and Interoperability

Awkward Array’s latest release, v2.9.0, focuses on what every production library needs: stability, performance, and smoother interoperability. This release brings improved C++ kernel performance and better CuPy/NumPy interoperability — fitting, given CuPy’s own major release this month.

Looking ahead, the Awkward Array team will be presenting at NVIDIA GTC in March, showcasing new developments in GPU-accelerated columnar analysis and Python/C++ interoperability. The community is also actively welcoming contributions in performance benchmarking, GPU backends, and documentation. See the contribution guide to get started.

Optuna v4.7: New Multi-Objective Samplers

Optuna has released v4.7, introducing two new multi-objective samplers and adding local hyperparameter importance computation to PedAnovaImportanceEvaluator. If you're doing hyperparameter optimization at scale, it's worth a look. Check the release notes for the full breakdown.

Open OnDemand 4.1

On the affiliated projects front, Open OnDemand — the popular open-source web portal for HPC access — has released version 4.1. This minor release builds on the project’s community-driven development approach and continues to simplify HPC workflows for researchers and institutions.

Read the full announcement: Open OnDemand 4.1 Release

That’s a wrap on February’s highlights! The NumFOCUS ecosystem continues to grow, evolve, and push the boundaries of open scientific computing. Want to see your project’s news featured here? Reach out to the NumFOCUS team with your updates.

As we begin a new year, we’re excited to share what’s ahead for NumFOCUS events in 2026. From the projects to PyData and SciPy, these conferences bring innovation and community together.

Our events are central to the NumFOCUS educational mission to bring together a global network of practitioners, researchers, maintainers, and learners to advance open, inclusive, and community-driven technology.

We’re grateful to the volunteers, speakers, organizers, sponsors, and attendees who make these community-driven events possible. From local meetups to global conferences, it’s all powered by you.

PyData Wrapped 2025: A Global Community in Motion

In 2025, NumFOCUS events brought the data science and scientific computing community together at an unprecedented scale:

• 216,966 members participated across our global PyData Meetup community to connect regularly at the local level
• 138 PyData Meetup groups served as community hubs for learning, collaboration, and leadership development
• Around the globe, 5,660 conference attendees came together at in-person and virtual PyData events
• Community members submitted 1,944 conference proposals, bringing fresh perspectives to the table for conferences
• 584 conference sessions were presented, spanning open source data tools, scientific Python, applied data science, and more
• PyData and NumFOCUS events were hosted in 11 cities across 5 countries, alongside one global online conference
• 68 countries were represented at PyData Global, highlighting the international reach of the community

From first-time speakers to community mainstays, PyData remains one of the best data science ecosystems for open collaboration, learning, and professional growth.

As we reflect on the milestones and momentum of 2025, we’re turning our focus to 2026 and the events we’re expecting in the year ahead:

• PyCon DE / PyData
• PyData London
• SciPy
• PyData Amsterdam
• PyData US
• PyData Tel Aviv
• PyData Eindhoven
• PyData Global

PyCon DE & PyData 2026

April 14–17, 2026 | Darmstadtium, Darmstadt (Frankfurt), Germany

PyCon DE & PyData is the largest gathering of the Python community in Europe! The three-day conference will feature hands-on workshops, informative keynotes, and educational talks, bringing together people from different backgrounds and skill levels to work and learn together.

April 14–16: Three days of talks, tutorials, and community connection across Python, data science, and open source tooling

April 17: A dedicated day of in-depth masterclasses

Tickets are available now. PyConDE & PyData tend to sell out, though, so you’ll want to snag yours soon.

PyData London 2026

June 5–7, 2026

PyData London continues to be one of the most influential PyData events on the calendar, bringing together a community across programming languages like Python, Julia, R, and more to learn and grow together. This three-day conference is perfect for data scientists, data engineers, and data analysis tool developers to share and learn from each other.

Call for Proposals open through Feb. 16

There’s still time to submit a talk or tutorial proposal! Learn more about how to submit a proposal here.

Talks v. Tutorials

Talks are 40-minute sessions with built-in Q&A times. Your proposal should include a short description that encourages someone to to come and learn about something.

Tutorials are 90-minute hands-on sessions where you would lead attendees through learning new skills, libraries, technologies, etc.

Volunteer applications close April 24

We are still accepting applications for volunteers to support PyData London 2026. Volunteer duties include working at the registration desk and monitoring talks, ensuring talks start and end on time, introducing speakers, moderating Q&A, and other needs as assigned. Learn more about how to apply here.

Tickets available now

You can get your tickets to PyData London 2026 right now! Grab them quick before they’re gone here.

SciPy 2026

July 13–19, 2026 | University of Minnesota | Minneapolis, MN

SciPy remains one of the longest-running and most respected open science collaboration events in the ecosystem. In 2026, the conference returns with a full week of tutorials, talks, and sprints. SciPy is where maintainers, researchers, and practitioners come together to push scientific Python forward — a flagship data engineering conference experience grounded in open source values.

Call For Proposals open now through Feb. 25

This is your chance to share innovative work on scientific computing with Python across a range of exciting tracks — including our two highlighted tracks: Spirit of SciPy and Data-Driven Discovery, Machine Learning and Artificial Intelligence — to domain-specific areas like physics, climate science, biology, education, and more. Submit your proposal now and help shape a week of learning, discovery, and code collaboration in 2026!

Introducing PyData US

We’re excited to announce PyData US, launching in Fall 2026.

This new conference will bring a large-scale PyData conference experience to the United States. Our goal is to host 500–600 attendees and create a vibrant space where ideas become action.

We are currently seeking a partner organization, university, or company to collaborate with us on a venue in:

• Washington, D.C.
• Baltimore
• New York City
• Boston
• Philadelphia

If you’re interested in hosting or have a connection to share, please contact Senior Events Manager, Tomara Youngblood, at tomara@numfocus.org.

More PyData Conferences Coming Soon

Additional PyData 2026 conference dates and information will be announced soon, including:

• PyData Amsterdam
• PyData Berlin
• PyData Tel Aviv
• PyData Eindhoven
• PyData Global — December 8–10, 2026 (online)

NumFOCUS Project Events

In addition to our PyData events, we’re excited to support the events our NumFOCUS projects are hosting:

BioCon (BioConductor)- August 10–12
JuliaCon 2026- August 10–15
Parsl- TBD (check back for more info soon!)
scverse- TBD (check back for more info soon!)

Sponsorship Opportunities

NumFOCUS events are made possible through the support of mission-aligned sponsors who believe in open source, open science, and inclusive innovation.

Sponsoring a PyData or NumFOCUS event is a meaningful way to:

• Support the sustainability of open source projects
• Engage with a global, highly technical audience
• Invest in inclusive tech events that prioritize community and impact

If your organization is interested in sponsoring an event, please reach out to Senior Events Manager, Tomara Youngblood, at tomara@numfocus.org.

From local meetups to global conferences, PyData is NumFOCUS — our educational arm bringing the open source community together to shape what’s next. We can’t wait to see you at a PyData conference in 2026.

In November, I told you that NumFOCUS was launching a strategic planning effort. The purpose of this work is to get to the heart of the organization and figure out what NumFOCUS needs to do not just to survive, but to thrive.

The exploration phase is well underway–thank you to everyone who completed the survey– invitations are going out to community members to participate in working sessions and interviews.

I also promised that we wouldn’t wait until the completion of the strategic planning process to make necessary changes at NumFOCUS.

To that end, we are diving in deep on all of our internal processes and policies. As we find issues, we are addressing them.

To get NumFOCUS back on solid financial ground, the Board of Directors approved a balanced budget for 2026. A budget that realistically captures expenses and eliminates the practice of budgeting to goal revenue.

But, being realistic about our revenue and expenses meant making the hard decision to reduce the size of the NumFOCUS team.

A smaller team combined with insights from our internal review also made it clear that we needed to reorganize our work.

As of February 1, the NumFOCUS staff will be organized in three departments–community, finance and operations and advancement.

Jim Weiss will lead the newly formed community department as Chief Community Officer, which incorporates staff from the project advocacy and events teams. In addition to serving as the lead point of contact for projects and organizer of events, the community department will also work to manage engagement, foster a sense of belonging and build and support networks. The goal is to more purposefully and strategically serve projects and the community.

With nearly a decade of service at NumFOCUS, Jim is well-known to the community. He most recently served as Director of Events and Resources and as Interim Executive Director. He has also served as a staff liaison and partner to numerous NumFOCUS committees–including leading the successful 2025 elections process. His deep network within projects and the community and clear understanding of the NumFOCUS mission and purpose, combined with his relationships across the staff team position him well to help shape this newly created department.

The development, fundraising and marketing teams will be housed under the newly created advancement department. And a search for a Chief Advancement Officer will take place in the second half of the year. This new unit will be responsible for strategically growing support for projects, the community and NumFOCUS operations. Shaping this strategy will be one of the key outputs from the strategic planning process.

Finally, I’m pleased to announce that Michael Savelli joins the team on February 1 as Chief Financial and Operations Officer. Michael will lead the newly combined finance and operations department with responsibility for finance, operations, HR, IT, legal and other related business functions.

Michael joins NumFOCUS with over 20 years of leadership experience guiding organizations like ours through times of growth, transition and transformation. His previous leadership roles include Chief Operating Officer and Interim Chief Financial Officer at the American Psychiatric Association, COO at the American Association for the Advancement of Science and most recently COO and CFO at the American Payroll Institute, where he led financial stabilization efforts and strengthened governance practices.

But Michael isn’t just a seasoned nonprofit executive, he brings a deep personal connection to open-source communities. His career began in software engineering and he remains an active Python developer, building automations to streamline operations and validate financial models, reinforcing his hands-on approach to continuous improvement.

I recognize that this is a lot of change for NumFOCUS in a short period of time. I am grateful to members of the community who shared their views with me and took time out to answer my questions and stress-test ideas. I very much appreciate the candor and thoughtfulness. And I know there is still much work ahead.

I look forward to continuing this work together through the strategic planning process. As always, please don’t hesitate to reach out in between scheduled sessions if there is anything you think I need to know.

Please join me in welcoming Michael to the team and in congratulating Jim on his promotion!

We’re kicking off the new year with a wide range of updates from across the NumFOCUS community. From new software releases and upcoming conferences to calls for participation and program announcements, our Sponsored and Affiliated Projects continue to move open science and data science forward. Here’s a roundup of the latest news.

Sponsored Project Updates

pandas

The pandas team has announced the release of pandas 3.0.0, a major milestone following the pandas 2.x series. This release brings together new features, performance improvements, and bug fixes, along with some intentional breaking changes that reflect the project’s ongoing evolution.

Key highlights of pandas 3.0 include making the dedicated string data type the default, introducing consistent copy/view behavior through Copy-on-Write (CoW) — effectively eliminating the long-standing SettingWithCopyWarning — and adopting a new default resolution for datetime-like data. The release also includes initial support for the new pd.col syntax, laying groundwork for future usability improvements.

As part of this major release, pandas 3.0 removes functionality that had already been deprecated in earlier versions. Users are strongly encouraged to first upgrade to pandas 2.3 and confirm that their code runs without warnings before moving to pandas 3.0. An overview of all potentially breaking changes is available in the “Backwards incompatible API changes” section of the release notes.

pandas 3.0.0 supports Python 3.11 and higher and is available via both PyPI and conda-forge. Users can install the release using:

PyPI:

python -m pip install — upgrade pandas==3.0.*

Or conda-forge:

conda install -c conda-forge pandas=3.0

The team encourages users to report any issues on the pandas issue tracker and extends a big thank-you to the many contributors who made this release possible.

ITK

The Insight Segmentation and Registration Toolkit (ITK) team has announced the release of ITK 5.4.5, a maintenance-focused update for this cross-platform, open-source toolkit supporting N-dimensional scientific image analysis with spatially-aware algorithms. This release emphasizes stability, documentation improvements, and platform support, while also introducing a new approach to onboarding AI agents to assist with project maintenance tasks.

On the bug-fix front, ITK 5.4.5 resolves issues related to zero-sized CompositeTransforms and out-of-bounds access in GDCM, addressing potential security vulnerabilities. The team also fixed bugs in imread when working with single-element lists and added new tests to further strengthen the reliability of image I/O.

Documentation received several meaningful updates, including the backporting of AnatomicalOrientation along with updated usage guidance. Additional refinements include spelling corrections, the removal of legacy documentation, and updates to shared guides. Notably, the new AGENTS.md file was introduced to help onboard AI agents by outlining project structure and contribution guidelines.

Platform support was also improved, with enhanced testing for macOS 15 (Intel), updates for Python with Xcode 16.2, and support for building against newer system versions of Eigen3.

Learn more about the release on GitHub: https://github.com/InsightSoftwareConsortium/ITK/releases/tag/v5.4.5

rOpenSci

rOpenSci has opened applications for the rOpenSci Champions Program 2026–2027, a 12-month initiative designed to grow open science and sustainable research software communities in Latin America, with a particular focus on Spanish-speaking participants.

The program is currently seeking both Champions and Mentors. Champions are R users and developers based in Latin America who are interested in building research software and strengthening local communities. Mentors are experienced R package developers and reviewers who want to support and guide emerging leaders in the open science ecosystem.

Key dates for the upcoming cohort include the call opening on January 12, 2026, a community call on January 21 where prospective applicants can meet past Champions and Mentors, an application clinic on February 5, and a final application deadline of February 20, 2026.

Participants in the program receive training workshops, one-on-one mentoring, opportunities for project development and community building, and a participation stipend. Full program details and application materials are available in both English and Spanish:

• English: https://ropensci.org/blog/2026/01/12/programchamps2026/
• Spanish: https://ropensci.org/es/blog/2026/01/12/programachamps2026/

rOpenSci also encourages community members to help spread the word across Mastodon and LinkedIn in both English and Spanish.

GeoPandas

The GeoPandas team has announced the release of geopandas v1.1.2. This release is now available for download and includes source distributions in multiple formats, making it easy for users and contributors to access and install the latest version.

You can find the full release details and download assets on the GeoPandas GitHub release page: https://github.com/geopandas/geopandas/releases/tag/v1.1.2

JuliaLang

The Julia community has shared several major updates, including both an upcoming conference and a new pre-release of the Julia language itself.

JuliaCon Global 2026 Call for Proposals is now open. JuliaCon Global 2026 will be held in person in Mainz, Germany, from August 10–15, 2026, and the Call for Proposals will close on February 28, 2026, at 23:59 CET. Remote presentations will not be possible for this event.

The JuliaCon organizing team invites proposals across a wide range of topics, from introductory to advanced, spanning academia and industry. Topics of interest include, but are not limited to, scientific computing, data analytics and visualization, numerical optimization, machine learning and AI, computational physics and chemistry, software engineering best practices, and applications in fields such as biology, finance, social science, and the humanities. The key criterion is whether a topic will be interesting and valuable to the Julia community.

Package maintainers are especially encouraged to submit posters about their packages, regardless of whether they are also submitting a talk. More information and the full Call for Proposals can be found on the JuliaCon website.

In addition, Julia version 1.13.0-beta1 — the first beta pre-release in the upcoming 1.13 series — is now available. This beta is intended for testing and feedback, particularly from package developers, and is not recommended for production use. Binaries are available via JuliaUp as well as manual downloads for macOS (Intel and Apple Silicon), Windows, Linux, and FreeBSD. Users can review the NEWS.md file and recent commits to see what’s new in this release series.

CuPy

The CuPy team has released the first release candidate for CuPy v14, currently planned for a full release in January 2026. The team is encouraging users to try out the release candidate and share feedback or report issues on GitHub.

Full release notes and details are available here: https://github.com/cupy/cupy/releases/v14.0.0rc1

Affiliated Project Updates

Taskflow

Taskflow has announced the official release of Taskflow v4.0, marking a major milestone for the project. Alongside the new release, the team has also launched Taskflow Academy, a new learning resource designed to help users get started and deepen their understanding of Taskflow.

• Taskflow v4.0: https://taskflow.github.io/
• Taskflow Academy: https://github.com/taskflow/academy

Project Pythia

Project Pythia has shared early details about the Pythia Cook-off 2026 Hackathon, which will take place in Boulder, Colorado, at the NCAR Mesa Laboratory. Registration is expected to open in late January or early February, and a limited amount of funding will be available to help support participant travel.

Additional details, including the full agenda and registration information, are available on the event website.

Optuna

The Optuna team released Optuna v4.7 on Jan. 19. In addition to this upcoming release, contributor Hiroaki Natsume has introduced a new optimization algorithm — SPEA-II (Strength Pareto Evolutionary Algorithm 2) — now available through OptunaHub.

More information about the SPEA-II sampler can be found here: https://hub.optuna.org/samplers/speaii/

We’re grateful to all of the maintainers, contributors, and community members who continue to make this work possible. Stay tuned for more updates from across the NumFOCUS ecosystem throughout the year.

Sponsored Projects

scikit-learn

scikit-learn is happy to announce the 1.8.0 release which you can install via pip or conda:
pip install -U scikit-learn
or
conda install -c conda-forge scikit-learn

You can read the release highlights under
https://scikit-learn.org/stable/auto_examples/release_highlights/plot_release_highlights_1_8_0.html, and the long list of the changes under
https://scikit-learn.org/stable/whats_new/v1.8.html.

This version supports Python versions 3.11 to 3.14 and features support
of free-threaded CPython 3.13 and 3.14. Thanks to all contributors who helped on this release!

astropy

We are very happy to announce the v7.2.0 release of astropy, a core Python package for Astronomy: http://www.astropy.org

The astropy core package is a community-driven Python package intended to contain much of the core functionality and common tools needed for astronomy and astrophysics. It is part of the Astropy Project, which aims to foster an ecosystem of interoperable astronomy packages for Python.

Notable changes in this release include:

• A faster ECSV table reader
• Generic DataFrame conversion methods
• Table index improvements and deprecation
• Cosmology traits
• An option to preserve units in FITS-WCS
• Concatenation and stacking of coordinates and time classes

In addition, hundreds of smaller improvements and fixes have been made. An overview of the changes is provided at: http://docs.astropy.org/en/stable/whatsnew/7.2.html

Instructions for installing astropy are provided on our website, and extensive documentation can be found at: http://docs.astropy.org

If you usually use pip to install packages, you can do:

pip install astropy — upgrade

If you make use of conda (such as through the Anaconda Python Distribution), you should soon be able update to Astropy v7.2.0 with:

conda update astropy

Or if you cannot wait for Anaconda to update their default version, you can use the conda-forge channel:

conda update -c conda-forge astropy

Please report any issues, or request new features via our GitHub repository: https://github.com/astropy/astropy/issues

Over 500 people have contributed code to the core astropy package so far, and you can find out more about the team here: https://www.astropy.org/team.html

If you use astropy directly for your work, or as a dependency to another package, please remember to acknowledge it by citing the appropriate Astropy paper. For the most up-to-date suggestions, see the acknowledgement page.

We hope that you enjoy using astropy as much as the team enjoyed developing it!

Cantera

On behalf of the Cantera development team, we are pleased to announce the availability of Cantera 3.2.0. There have been nearly 1000 commits to Cantera since the last version, 3.1.0, which was released in December 2024. They have closed or merged 144 pull requests and closed 67 issues and 19 enhancement proposals. You can install Cantera 3.2 now using Conda (conda-forge), Pip, or Apt (Ubuntu).

This release was made possible with the help of 17 fabulous contributors! The full release notes for Cantera 3.2.0 are given here.

pandas

We are happy to announce the first release candidate of pandas 3.0.0, released last week.

You can find the the list of changes in 3.0.0 in the release notes page.

Users having pandas code in production and maintainers of libraries with pandas as a dependency are strongly recommended to run their test suites with the release candidate, and report any breaking change to our issue tracker before the official 3.0.0 release.

Especially the changes around the new default string dtype and Copy-on-Write are expected to require some code changes. See the linked release notes for details how to test and update your code.

The release candidate is available on PyPI and conda-forge, for example:

python -m pip install — upgrade pandas==3.0.0rc0
or
conda install -c conda-forge/label/pandas_rc pandas==3.0.0rc0

rOpenSci

December was an exciting month for rOpenSci!

🌎 rOpenSci proudly continued supporting LatinR as a community partner in 2025. Here they share a list of resources and recordings for the tutorials and talks delivered by their staff and community memebers at LatinR. Discover more on the LatinR YouTube channel.
5️⃣ rOpenSci’s community manager, Yani (Yanina Bellini Saibene), was recently interviewed on our new NumFOCUS podcast, Give Me 5. You can watch the 5-minute episode now. Yani’s central message: people are the heart of open source. And when we invest in those people — maintainers, contributors, educators, and organizers — we strengthen the global scientific community.
🤝Two new Coworking sessions have been revealed:
- Tuesday January 13th, 9:00 Americas Pacific (17:00 UTC), “Let it go!” with Steffi LaZerte and cohost Yanina Bellini Saibene.
- Tuesday February 2nd, 9:00 Australia Western (01:00 UTC), “Share your Positron setup!” with Steffi LaZerte and cohost Noam Ross.
📦 The mantis package, developed by T. Phuong Quan, recently became a part of rOpenSci’s software suite

Learn more on rOpenSci’s blog!

Affiliated Projects

Optuna

Optuna published Optuna Dashboard Integrates with LLM: Enabling Data Filtering and Graph Generation using Natural Language. This article explains how the recently introduced Optuna Dashboard LLM integration feature works and is implemented. Additionally, in OptunaHub, Noisy Expected Improvement for RobustGPSampler is supported.

MFEM

Version 4.9 of MFEM was just released on December 11, 2025. Highlights of this release include:

• Introducing ∂FEM: a new MFEM capability for Automatic Differentiation (AD) of nonlinear finite element operators, based on Enzyme or dual numbers AD at quadrature points. These features are part of the new mfem::future namespace. See the new fem/dfem/ directory and the minimal surface miniapp for illustration of ∂FEM’s use.
• Initial support for particle methods in MFEM with new classes Particle, ParticleSet and ParticleVector.
• Added a new miniapp and specialized AMGF solver for optimization-based contact mechanics. The miniapp solves large-scale frictionless contact using aself-contained Interior Point (IP) solver, mortar-based contact constraints provided by Tribol. For more details, see the miniapps/contact directory.

This release contains many other improvements, all of which have been summarized here!

Skforecast

Skforecast 0.19.0 has just been released! 🎉

This release expands skforecast into new types of forecasting problems, strengthens drift detection, and introduces a dedicated statistical modeling module fully aligned with the scikit-learn interface.

🎯 Autoregressive Classification Forecasting

Meet the new ForecasterRecursiveClassifier, designed for time series classification tasks where the goal is to predict future class labels instead of numeric values. Perfect for scenarios where understanding what category comes next matters more than the exact value.

🧭 Detecting Data Drift with Confidence

The new PopulationDriftDetector helps you identify when your data distribution changes over time, signaling that your model may need retraining. It supports both target and exogenous variables, in single and multiseries forecasting — ideal for monitoring production pipelines.

📊 New Statistical Models Module
We’re introducing a new stats module that brings classic time series models into skforecast with a simple, unified interface. It includes the new Arar model, an algorithm that combines memory-shortening transformations with autoregressive modeling.

With this release, skforecast continues to bridge the gap between modern ML forecasting and classical time series analysis, offering even more flexibility for research and production.

🔗 Full details: https://skforecast.org/0.19.0/releases/releases
🔗 Documentation: https://skforecast.org/

From the NumFOCUS Election Committee: Alan Lujan, Ianna Osborne, Jacob Schreiber, Andrea Gomez Vargas, Mridul Seth, Rachel Kerestes

After a thorough process — with 76 percent of registered voters casting ballots–we are pleased to announce the results of the 2025 NumFOCUS Board of Directors election.

Noor Aftab, James A. Bednar and Francesca van Doorn were elected by the community to serve a two-year term on the Board of Directors, with the option to renew for an additional term. Their terms will begin in January 2026.

Voting results were reviewed and verified by both the Election Committee and the current NumFOCUS Board of Directors. All candidates were notified of the results of the voting on December 12. The winners each formally accepted the position.

Ballots were cast by registered voters which included representatives of sponsored projects, affiliated projects, NumFOCUS committee members, event chairs and current members of the Board of Directors. This breadth of participation helps ensure that the Board of Directors reflects the diverse voices and experiences that make up the NumFOCUS community.

We extend our sincere thanks to everyone who participated in this year’s election, including those who submitted nominations, applied to serve, registered to vote and cast ballots. Your engagement is essential to maintaining NumFOCUS as a community-centered organization.

Please join us in congratulating James A. Bednar, Noor Aftab, and Francesca van Doorn on their election to the NumFOCUS Board of Directors. We look forward to working together in the years ahead as NumFOCUS continues to advance open source science for the public good.

Biography:

Noor Aftab

Noor Aftab is a leader in Scientific Computing and Responsible AI, currently serving as a Senior Program Manager at Amazon S3. A specialist in governance and ecosystem integrity, she previously chaired the NumFOCUS Inaugural Code of Conduct Working Group. Noor has been appointed to the American Statistical Association’s Committee on Data Science and AI (2026) and is the author of “Unlocking the Missing 78%,” research addressing bias and fairness in large language model (LLM) ecosystems. A frequent keynote speaker at conferences such as PyData Global and IEEE, she is committed to the stewardship and ethical evolution of the scientific computing stack.

Francesca van Doorn

Francesca van Doorn is the Director of Finance and Operations at Windward Fund, a nonprofit fiscal sponsor supporting environmental science and advocacy projects. In this role, she oversees financial operations and provides strategic guidance for a large and complex portfolio of fiscally sponsored initiatives. Previously, Francesca served as Financial Services Director at Arabella Advisors, where she supported nonprofit and philanthropic organizations through outsourced operational and financial services. She brings deep expertise in nonprofit financial management, grants administration, compliance, and governance across both 501(c)(3) and 501(c)(6) organizations. Francesca’s experience managing large fiscal sponsor organizations and navigating the financial, legal, and ethical considerations of mission-driven work positions her to contribute meaningfully to NumFOCUS’s financial and operational leadership.

Dr. James A. Bednar

Dr. James A. Bednar is the Director of Professional Services at Anaconda, Inc., leading a team of developers supporting a suite of open source scientific software projects. He founded the HoloViz project, adopted as a fiscally sponsored NumFOCUS project in 2023, and manages its Datashader, Param, and Colorcet packages. Previously, he was a professor at the University of Edinburgh, where he led the Doctoral Training Centre for Neuroinformatics and published over 50 scientific papers. James has over 20 years of experience on academic journal review boards and has been active in the SciPy community for more than 15 years, contributing through reviewing, speaking, and tutorials. His extensive experience in founding, leading, and managing open source projects and academic initiatives equips him to bring strategic and technical insight to the NumFOCUS Board of Directors.