PyData Paris 2025
| title | author | length |
|---|---|---|
| Big ideas shaping scientific Python: the quest for performance and usability | Ralf Gommers | 47m |
NumPy 2.0 lands with a June 16 rollout that requires recompiling downstream wheels, tracks ~100 packages and takes roughly a year and a half for ecosystem… | ||
| You Don’t Have to Be an Expert: Stories from the Open Source Frontlines | Alenka Frim | 35m |
A newcomer with no PyArrow experience four years ago describes concrete wins: a proof of concept at Prague sprints that leads to a PR merged this month folding… | ||
| Building Data Science Tools for Sustainable Transformation | Anita Graser | 33m |
Argues that turning generative AI hype into durable tools means shipping concrete integrations and workflows not slogans: movingpandas is an OSG community… | ||
| Open-source Business | Sylvain Corlay, Yann Lechelle, Alexander CS Hendorf | 32m |
Claims about funding and governance hit specifics: a €80M Series A example, government grants seeding projects, and companies of ~30 employees balancing paid… | ||
| Advanced Polars: Lazy Queries and Streaming Mode | Emanuele Fabbiani | 28m |
Polars claims Rust-level speed and Apache Arrow memory layout to run lazy queries that optimize across steps using predicate and projection pushdown and… | ||
| A Hitchhiker's Guide to the Array API Standard Ecosystem | Lucas Colley | 28m |
The project claims a practical unification of Python array libraries so NumPy CuPy PyTorch JAX and Dask interoperate; it ships a test suite and wrappers and… | ||
| A Journey Through a Geospatial Data Pipeline: From Raw Coordinates to Actionable Insights | F Gravin | 25m |
Pipeline processes a 3‑day Swiss rainfall raster set: download COGs, stack into an Xarray cube, clip to the Swiss boundary, run per-tile filters and a… | ||
| Applying Causal Inference in Industry 4.0: A Case Study from Glasswool Production | S Bottani, P Lee | 33m |
Case study from Saint-Gobain with Inria claims a measurable cut in raw material use reduced raw material usage without compromising product quality and shows… | ||
| Advancements in optimizing ML Inference at CERN | Sanjiban Sengupta | 27m |
SOFIE converts Keras/PyTorch/ONNX models into standalone C++ plus weight files with only a BLAS dependency, claiming support for ~75–80 operators and idiomatic… | ||
| How to make public data more accessible with "baked" data and DuckDB | Chris Kucharczyk | 29m |
Bakes IPEDS into a portable DuckDB file so analysis runs locally or on a tiny web worker, produces about 22 million rows in a 320 MB relational DB, and exposes… | ||
| Beyond Prototyping: Building Production-Level Apps with Streamlit | Johannes Rieke, Arnaud Miribel | 28m |
Streamlit claims it scales from prototype to production: Snowflake reports 500 internal apps across 70 teams using templates, preview deployments on PRs, and… | ||
| Code as Data_ A Practical Introduction to Python’s Abstract Syntax Tree | Laurent Direr | 32m |
Shows a concrete claim: Python’s ast lets code be rewritten so pytest prints intermediate values instead of a bare AssertionError, by rewrites assert into… | ||
| Browser-based AI workflows in Jupyter | Jeremy Tuloup, Nicolas Brichet | 31m |
Extension of JupyterLite brings Python into the browser and adds an AI stack: a chat widget with tool calls, an agent discovers 437 Jupyter commands and can… | ||
| Building Resilient (ML) Pipelines for MLOps | Lex Avstreikh | 33m |
Claims a platform approach works where ad hoc boxes fail and shows a concrete architecture: feature training inference split, a metadata view abstraction with… | ||
| ActiveTigger: A Collaborative Text Annotation Research Tool for Comp. Social Sciences | E. Schultz | 24m |
ActiveTigger is a working open source annotation stack: a Python backend and React frontend that computes embeddings, offers a collaborative (asynchronous)… | ||
| From Jupyter Notebook to Publish-Ready Report: Effortless Sharing with Quarto | Christophe Dervieux | 28m |
Quarto converts a Jupyter notebook into a shareable site or PDF with a single command — quarto publish — and uses Pandoc under the hood to produce multi‑format… | ||
| How to do real TDD in data science? A journey from pandas to polars with pelage! | A. Tiran-Cappello | 32m |
Claims that Pelage makes real TDD feasible in data workflows and that its Polars integration stays lightweight. The demo builds an open‑source testing layer on… | ||
| State of Parquet 2025: Structure, Optimizations, and Recent Innovations | Rok Mihevc, Raúl Cumplido | 31m |
Format details: file layout with row groups, column chunks, pages and footer; page indexes and Bloom filters let readers skip pages by min/max and… | ||
| Enhancing Machine Learning Workflows with skore | Marie Sacksick | 28m |
skore presents a concrete ML workflow layer on top of scikitlearn and scrap, ships an estimator report that trains and stores models, a comparison reports… | ||
| Balancing Privacy & Utility: Efficient PII Detection and Replacement in Textual Data | Justine Bel-Letoile, Elizaveta Clouet | 40m |
Pipeline claims a practical, open‑source two‑step solution: detect PII with pre‑trained NER (gliner) then replace with controlled synthetic data via Faker… | ||
| Move beyond academia: Introducing an industry-first tabular benchmark | Alexandre Abraham | 31m |
A benchmark and a pipeline package that run complex pre/postprocessing on real industry problems and synthetic proxies, built from private industry datasets… | ||
| Prediction intervals | Olivier Grisel | 34m |
Covers concrete methods and code: training ensemble of quantile regressors (scikit‑learn, XGBoost, PyTorch), reducing probabilistic regression to binning +… | ||
| Tackling Domain Shift with SKADA: A Hands-On Guide to Domain Adaptation | Théo Gnassounou | 31m |
SKADA is a Python library that brings domain adaptation into scikit-learn and PyTorch integration, providing ~20 shallow methods plus deep approaches (e.g… | ||
| Modern Web Data Extraction: Techniques, Tools, Legal and Ethical Considerations | Domagoj Marić | 35m |
Clear claims and concrete examples: which site yielded what fields, how many records, and how long the crawl takes; which exact tools win (BeautifulSoup +… | ||
| Architecting Scalable Multi-Modal Video Search | Sebastiano Milardo, Pietro Piccini | 33m |
An end-to-end system indexes petabytes by extracting tracked crops, embedding them and storing vectors for fast nearest-neighbor search; preprocessing uses… | ||
| Build a data studio in your notebook with jupyter-fs | Tim Paine | 28m |
jupyter-fs maps the Jupyter contents API to PyFilesystem and fsspec backends and ships as a contents manager plus server and JupyterLab front-end extensions… | ||
| Meta-Dashboards: Accelerating Geospatial Web Apps Creation with Voilà | Davide De Marchi | 30m |
The Joint Research Centre open-sources a Voila simplification library that composes ipywidgets, ipyvuetify, ipyleaflet, Plotly and bespoke SVG charts to… | ||
| Optimal Transport in Python: A Practical Introduction with POT | Rémi Flamary | 29m |
Claims practical Python code that solves OT problems with concrete scaling: exact solvers run about n^3 log n so handle a few thousand to tens of thousands of… | ||
| CoSApp: an open-source library to design complex systems | Étienne Lac | 26m |
CoSApp is an open‑source Python library (docs on Read the Docs, code on GitLab) that actually assembles models and solvers: explicit ports with automatic unit… | ||
| Documents Meet LLMs: Tales from the Trenches | Nour El Mawass, Miklos Erdelyi | 32m |
Pipeline extracts marketing claims from brochures, uses OCR (Amazon textract, Azure AI document intelligence or mark it down) and prompts for XML output or… | ||
| CodeCommons_ Towards transparent, richer and sustainable datasets for code generation model training | Simeon Carstens, Rania Talbi | 28m |
CodeCommons builds a metadata-rich code corpus on top of Software Heritage and claims an ingestion pipeline that will catch up ~150 million GitHub… | ||
| Collaborative GIS editing in JupyterLab | Martin Renou, Arjun Verma | 28m |
JupyterGIS claims live collaborative editing of GIS files inside JupyterLab, including supports QGIS project import/export, with a programmatic Python API that… | ||
| Parallel processing using CRDTs | David Brochart | 28m |
Claims that CRDTs let subprocess workers mutate local copies with automatic merge and low latency, demonstrated using PICRDT (wires/Yjs binding) and contrasted… | ||
| Sparrow, Pirates of the Apache Arrow | Johan Mabille, Alexis Placet | 23m |
A lightweight C++20 implementation of the Apache Arrow memory spec claims practical compatibility with the Arrow C data interface, zero dependencies (except… | ||
| Reproducible software provisioning for high performance computing & research software engineering | Martin Lang, Hans Fangohr | 26m |
Institute runs Spack package manager to reproducibly build and expose a full HPC stack via environment modules, with 600 packages across six architectures… | ||
| Sharing computational course material at larger scale: a French multi-tenant attempt | N. M. Thiéry | 30m |
A French multi-tenant project produces a shared corpus of modular Python training units claiming FAIR reuse rather than full courses, built as markdown/Jupyter… | ||
| Repetita Non Iuvant: Why Generative AI Models Cannot Feed Themselves | Valeria Zuccoli | 30m |
Claim: feeding models their own outputs progressively stable diffusion trained with 70% real 25% generated collapses after five retrains and OPT shows… | ||
| Machine Learning in the Browser: Fast Iteration with ONNX & WebAssembly | Romain Clement | 33m |
Claims that shipping edge demos fixes the integration gap where up to 80% of models never reach product; shows concrete pipeline: train in Python, export via… | ||
| Expanding Programming Language Support in JupyterLite | T Beier, I Thomas, I Paredes, A Prouvost | 29m |
emscripten-forge provides a conda-style WebAssembly distro that compiles C/C++/Fortran into wasm32 and powers JupyterLite kernels so xeus-Python, xeus-R, and… | ||
| Navigating the security compliance maze of an ML service | Uwe L. Korn | 30m |
Practical run-through of compliance chores with concrete tools and outcomes: ISO-27001 certification completed, discussion of sift/trivy scans, generation of… | ||
| Fighting against the instability : Debian Science at the synchrotron SOLEIL | Emmanuel Farhi | 33m |
Facility produces about 3 PB/year of raw synchrotron data across 29 beamlines, serves ~4,000 visitors, and faces a dependency forest: 54 core apps pull ~2,000… | ||
| Skrub: machine learning for dataframes | (unnamed presenter) | 26m |
Skrub claims to convert Pandas/Polars tables into scikit‑learn ready feature matrices with minimal code and reproducible operations; it supplies concrete… | ||
| PyPI in the face: running jokes that PyPI download stats can play on you | Loïc Estève | 29m |
Analysis uses the public bigquery dataset of PyPI logs and concrete tools like Pepy and pip stats to show which requests inflate counts: mirrors, request-based… | ||
| The new lockfile format introduced in PEP 751 | Nico Albers | 29m |
PEP 751 (accepted March 2025) specifies a common lockfile, pylock.toml format, that lists precise packages, indices, Python version and per-distribution URLs… | ||
| torchFastText: Modernizing Text Classification at Insee with PyTorch-based models | Tayebjee, Couralet | 32m |
Open-sourced torchFastText replaces archived fastText for INSEE by providing a PyTorch training stack that trains on GPU but runs CPU-only inference, targets… | ||
| xeus-cpp, the new C++ kernel for Jupyter. | Johan Mabille, Anutosh Bhat | 30m |
Claims a full replacement for xeus-cling that runs modern C++20/23, built as a plugin-friendly kernel exposing LLVM/clang rippling plus a new WebAssembly… | ||
| Unlock the full predictive power of your multi-table data | Luc-Aurélien Gauthier, Alexis Bondu | 32m |
Claims an automated, supervised feature-engineering system for raw multi-table data and ships as open-source Python library KOPS. Describes an MDL-based… | ||
| You Don’t Need Spark for That: Pythonic Data Lakehouse Workflows | Romain Clement | 33m |
Claims that many lakehouse tasks don’t need Spark for routine work and that modern Python stacks handle them: use delta lake Python library… | ||
| Lightning talks - session 1 | 53m | |
Choices hinge on specifics: Python native interface proposes a new C API and universal ABI with a PEP and demos claiming Python can beat Cython for some… | ||
| Lightning talks - session 2 | 58m | |
Quick patrol of specifics: NumFOCUS says it sponsors 65 projects and the channel is at 169,000 subscribers aiming for 170k; EU chat control claims automatic… | ||
| Backstage interview | Matthew Topol | 09m |
Apache Arrow's concrete claims and ecosystem work matter here: a memory layout that is identical on wire and in memory enabling zero copy communication… | ||
| Backstage interview | Anita Graser | 07m |
History of QGIS moves from 15–20 person developer meetings to 400–500 tickets sold out user conferences, driven by a deliberate sustainability plan and the… | ||
| Backstage interview | Auguste Baum, Ivan | 07m |
Says a company built its product on top of the scikit loan ecosystem and delivers an open-source library called Score that reduces boilerplate for model work… | ||
| Backstage interview | Florian Praden | 06m |
CFM describes how Python is driving production ML: models built with scikit learn feed a layered decision system that turns predictions into trade executions… | ||
| Backstage interview | Ian Thomas | 05m |
Describes adding multi-threaded Jupyter kernels support and the low-level async socket work required, explains Jupyter Lite compiled to WebAssembly and the… | ||
| Backstage interview | Jim Dowling | 05m |
A European full‑stack AI platform called Hopsworks claims to provide a feature store and real‑time LLM serving that runs on Kubernetes, is deployable with helm… | ||
| Backstage interview | Lucas Colley | 05m |
Video reports a concrete push to unify array libraries via the array API standard and an existing cross‑ecosystem consortium that spent years designing it… | ||
| Backstage interview | Nour El Mawass, Maria Knorps | 07m |
Generative pipeline moves from POC to production: deployed on AWS and stitches LLMs, RAG and vector search with post-filtering and retry logic to handle noisy… | ||
| Backstage interview | Raúl Cumplido, Rok Mihevc | 08m |
Overview of concrete outcomes and tools: Apache Parquet as a columnar on‑disk format supports encryption, a new content defined chunking feature in the C++… | ||
| Backstage interview | Simeon Carstens | 06m |
A project builds a large dataset for code‑generation LLMs on top of Software Heritage, ingesting all versions of open source repositories and running on an HPC… | ||
| Backstage interview | Simona Bottani | 05m |
Walks through an applied causal-inference pipeline for a manufacturing line: data collection from sensors, building a causal graph from expert knowledge, and… | ||
| Backstage interview | Cheuk Ting Ho | 05m |
Developer advocate for JetBrains and a member of the board of directors of the Python Software Foundation, focused on conference outreach, blog posts, and… | ||
| Backstage interview | Yann Lechelle | 07m |
A company forms as a joint venture to commercialize scikit‑learn while keeping the core open and BSD3 license code public, aiming for sustainability by… | ||
| Backstage interview | Nathalie Moreau David | 02m |
One-year tenure at CFM coordinates deliveries across multiple machine learning and AI projects while teams build an internal AI platform for production… | ||
| nothing matches | ||