PyData Amsterdam 2025
2025-09-24..2025-09-26
Amsterdam, Netherlands
[PyData Amsterdam 2016, PyData Amsterdam 2017, PyData Amsterdam 2018, PyData Amsterdam 2019, PyData Amsterdam 2024, PyData Amsterdam 2025]
| title | author | length |
|---|---|---|
| 10-year anniversary | Aftermovie | 02m | |
A celebratory aftermovie that documents 10-year anniversary milestones with footage of sessions social events and stage moments, showing over a thousand… | ||
| Help! There Are Humans in My Data! | Marysia Winkels, Isabelle Donatz-Fest | 31m |
Fieldwork shows concrete costs: 350 hours of shadowing police, police systems with 685 incident codes, downstream use in predictive policing, and the Titanic… | ||
| Kickstart Your Probabilistic Forecasting | Inge van den Ende | 33m |
Introduces two practical probabilistic forecasting methods — the Level Set Forecaster and Quantile Regression Forest — and explains how they provide… | ||
| Searching for My Next Chart | Muhammad Chenariyan Nakhaee | 34m |
Builds a RAG-based retrieval system that turns scattered ggplot2 and Datawrapper visuals into a semantically searchable corpus, embedding plot code and… | ||
| Lightning Talks | 22m | |
Rapid five-minute demos surface concrete claims and results: a personal anecdote links Stannislav Petrov prevented nuclear war to rescue-cat photos; a Sphinx… | ||
| Is Prompt Engineering Dead? | Iryna Kondrashchenko, Oleh Kostromin | 30m |
Automatic prompt optimization claims to replace manual tuning by treating prompts as trainable parameters and ships working frameworks: DSPy, AdelFlow, and… | ||
| Real-Time Context Engineering for LLMs | Manu Joseph | 24m |
Includes a working TikTok clone wired to a Hopsworks feature store and application IDs that unlock per-application context for agents, showing a concrete… | ||
| KEYNOTE: Data Architecture Turned Upside Down | Hannes Mühleisen | 37m |
Claim: flip the stack so compute moves to clients while storage and metadata stay commoditised. Concrete evidence: count distinct on a billion rows in 46s on a… | ||
| Orchestrating success | Rodrigo Loredo, Oscar Ligthart | 35m |
Vinted coordinates over 20 decentralized data teams handling petabytes by building a Python generator on top of Apache Airflow that renders DAGs from a simple… | ||
| Resource Monitoring and Optimization with Metaflow | Gergely Daroczi | 35m |
Metaflow integration adds a lightweight tracker that monitors CPU memory GPU VRAM network and disk at process and system level and claims a single line of code… | ||
| Sieves - Plug-and-Play NLP Pipelines With Zero-Shot Models | Raphael Mitsch | 31m |
A modular toolkit claims to speed prototyping of document processing by composing small tasks around zero-shot-first models and enforcing structured outputs… | ||
| Techie vs Comic - The sequel | Arda Kaygan | 17m |
A comedian who also does data science chronicles concrete moments: comparing transformer milestones to personal timelines, running LLMs and GenAI onstage as… | ||
| Optimize the Right Thing | Shimanto Rahman | 48m |
An open-source Python tool Empulse package integrates with Scikit-Learn and popular boosters (XGBoost CatBoost LightGBM) to translate a domain cost matrix into… | ||
| Scaling Trust | George Chouliaras, Antonio Castelli | 34m |
Claim: replace most large-scale human labeling with an LLM-based evaluator trained to match annotator judgments. The process builds a golden data set with… | ||
| Untitled13.ipynb | Vincent D Warmerdam | 49m |
Shows a practical rethinking of notebooks: code cells become reactive DAG nodes with immutable globals, updates propagate automatically, and deletion triggers… | ||
| Evaluating the alignment of LLMs | Iva Gornishka, Laurens Samson | 30m |
City of Amsterdam builds a public assessment platform with tailored Dutch benchmarks and a model table that reports concrete claims: CodeCarbon energy… | ||
| Optimal Observability | Vitalie Spinu | 34m |
Presents a concrete pipeline: define a BDR/BDP detection metric that accounts for counts, use inverse-harmonic-mean ordering plus moving-window agglomeration… | ||
| How to Keep Your LLM Chatbots Real - A Metrics Survival Guide | Maria Bader | 24m |
Concrete evaluation recipe: build a synthetic test set from Chroma-embedded docs and run a retriever-augmented-generation stack backed by GPD4.1 and Molly GPT… | ||
| Kafka Internals I Wish I Knew Sooner | Dima Baranetskyi | 35m |
Kafka reclaims disk only when segments roll so a one‑day retention with huge segment size effectively never frees space; default retention is seven days and… | ||
| Declarative Feature Engineering | Miguel Leite, Vitalii Zhebrakovskyi | 36m |
Adyen builds a Python-based DSL that declaratively defines features and auto-generates Spark and Flink pipelines, metadata, Airflow DAGs and Cassandra/Delta… | ||
| KEYNOTE: Ethics is Not a Feature | Dr. Maria Börner | 41m |
Claim: ethics cannot be tacked on as a feature but must shape data, modeling and deployment decisions. Examples and tools: image generators (Stable Diffusion… | ||
| Quiet on Set - Building an On-Air Sign with Open Source Tech | Danica Fine | 40m |
Builds a working on‑air sign using a Raspberry Pi that lights within a second when camera logs show AV capture or a Zoom port check indicates a meeting, with… | ||
| The Gentle Monorepo - Ship Faster and Collaborate Better | Gerben Dekker | 35m |
Dexter migrates incrementally from separate repos to a single Python monorepo by starting with fat bases, sharing one root pyproject.toml and a locked… | ||
| KEYNOTE: Image processing, AI, and autonomous systems | Judith Dijk | 41m |
Overview of image-processing tasks with concrete solutions and failure modes: ship detection in harbors (example contrasts MS COCO (80k) pretraining with 1,600… | ||
| Designing tests for ML libraries | Sayak Paul, Benjamin Bossan | 50m |
Shows concrete testing practices from HuggingFace: how diffusers and PFT run a stratified test schedule (fast per-PR, slow on main, nightly GPU runs)… | ||
| Applying Privacy-Enhancing Technologies | Yuliya Sapega, Joanna Pasiarska | 38m |
Hands-on tests compare real Python packages and show what actually works: federated learning stacks perform (Flower is easiest to deploy, PySyft suits… | ||
| No labels? No problem! | Jaap Stefels, Itzel Belderbos | 35m |
A production ML pipeline that labels fraud without true labels uses a two-step proxy labels system: first find suspect intervals with temporal anomaly… | ||
| Context is King | Bauke Brenninkmeijer | 49m |
Argues whether Retrieval-Augmented Generation stays necessary versus dumping documents into huge windows, and shows concrete numbers: ~100k tokens is a… | ||
| Monitoring of model drift in Finance | Denis Gaitan, Agustin Iniguez | 35m |
A deploy-gate system at Rabbo Bank automates model-drift checks by building draft states and comparing them with Azure DevOps pipelines, orchestrated via… | ||
| Fast-Track Your AI-ML Model Integration | Florenz Hollebrandse | 34m |
Claims a practical MDLC that starts productization day one and often cuts time-to-market by 50%. Shows two concrete projects (J.P. Morgan alerting pipeline and… | ||
| Time Series Decomposition with PyMC | Wesley Boelrijk | 36m |
Interpretable time‑series decomposition built with PyMC separates trend, seasonality and noise using engineered features and actually fits: a linear slope… | ||
| GenAI governance in practice | Maarten de Ruiter | 36m |
Claims that only 5% ROI of GenAI pilots succeed, shows up in a small data quality experiment that compares 10–50 documents using similarity searches and… | ||
| Causal Inference Framework for incrementality | Netesh, Nazlı Alagöz | 33m |
Estimates the incremental effect of app installs on one‑year booking value using propensity‑score matching, inverse‑propensity weighting, doubly robust /… | ||
| Counting Groceries with Computer Vision | Sven Arends | 31m |
Automated stock counting runs on 16 camera rigs feeding a FastAPI service and an LLM proxy, logging predictions to Snowflake and sampling a small percent for… | ||
| Uncertainty Unleashed | Konstantinos Tsoumas | 35m |
Claims that conformal prediction gives provable marginal 90% coverage on finite data and is model‑agnostic; demonstration wraps a classifier on MNIST using… | ||
| Large-Scale Video Intelligence | Irene Donato, Antonino Ingargiola | 35m |
Claims a scalable multi‑modal video search by embedding detected entities rather than whole videos and demonstrates it works at scale. Batch pipeline uses YOLO… | ||
| Should Captain America Still Host Your Data? | Manuel Spierenburg | 35m |
US law (the US Cloud Act access) can legally compel American clouds to hand over data; cited incidents include the 2022 Amsterdam Trade Bank cutoff and… | ||
| KEYNOTE: The agentification of software has a UX problem | Demetrios Brinkmann | 50m |
Argues that chat-based agents overpromise and underdeliver, citing loss of high‑bandwidth context, the prompting dance, uncertain attribution, and a real… | ||
| Model Context Protocol | Fabio Lipreri, Gabriele Orlandi | 52m |
Anthropic’s MCP claims a universal, bidirectional interface that turns APIs, databases, Slack, GitHub and local files into first‑class context providers; the… | ||
| nothing matches | ||