PyCon DE & PyData 2026
– · darmstadtium, Darmstadt
PyCon DE & PyData 2026 took place 14–17 April 2026 at the darmstadtium in Darmstadt: 145 sessions across 20 tracks, from 165 speakers.
Every session has a page here, with its abstract, speakers and — where a recording exists — the video. The full programme is below — filter it by track, format and level.
Descriptions and previews are generated by Open-Source AI.
-
Recording available. Catch the LLM if you Can: Watermarking LLMs
Large Language Model (LLM) watermarking addresses the challenge of distinguishing AI-generated text from human-written content to combat scams and prevent model collapse caused...
-
Recording available. Offline Fallback for a Mobile LoRaWAN Gateway
LoRaWAN networks typically rely on a cloud-based network server for decryption and data decoding, creating a critical dependency on internet connectivity. In disaster...
-
Recording available. pytest tips and tricks for a better testsuite
Pytest is a mature Python testing framework that provides advanced tools for managing complex test suites, focusing on strictness, parameterization, and efficient resource...
-
Recording available. In Praise of Documentation: Tools, Tips & Techniques for Literate Programming in the AI Age
Poor documentation creates institutional memory loss, making it difficult for new engineers to understand legacy codebases and leading to expensive, time-consuming...
-
Recording available. Hierarchical Models in MMM: Can Structure beat data size?
Marketing Mix Modeling (MMM) often faces the challenge of insufficient data for specific regions, where the number of available data points is equal to or less than the number...
-
Recording available. Building Agentic Systems with Python, LangGraph, MCP, and A2A
Agentic systems can be constructed using a hub-and-spoke architecture to automate complex business processes, such as identifying and scoring companies for mergers and...
-
Recording available. Django-Q2: Async Tasks Made Simple
Synchronous task execution in Django applications can lead to poor user experiences and system instability. When a web request triggers a slow or unreliable process, such as...
-
Recording available. Increase productivity of CNC-machining of aerospace engine parts with Python
CNC machining of aerospace engine parts, such as turbine shafts and cases, faces productivity challenges due to strict qualification processes, high-cost specialized machinery,...
-
Recording available. Causal Inference through the lens of probabilistic programming
Causal inference aims to determine the effect of interventions, such as price changes or marketing campaigns, by estimating the difference between an observed outcome and a...
-
Recording available. From Struggling to Mastery: A Practical Guide to Data Pipeline Operations
In "From Struggling to Mastery: A Practical Guide to Data Pipeline Operations," Akif Cakir, a seasoned Data and AI enthusiast with over 14 years of experience in the data...
-
Recording available. Rediscovering single-node processing: When does it make sense to move from Spark to Polars?
Apache Spark and Polars represent two different philosophies of data processing: horizontal scaling via distributed clusters and vertical scaling via single-node optimization....
-
Recording available. AI Evals Done Right: From Vibes to Confident Decisions
The provided transcript contains no technical content, data, or descriptions of methodologies regarding AI evaluations. It consists entirely of repeated expressions of...
-
Recording available. Using Sensor Fusion and ML to Navigate Underground When GPS Fails
Underground navigation is challenging because GPS signals cannot penetrate subway tunnels, leaving smartphones to rely on imprecise cell tower mapping or Wi-Fi scanning with...
-
Recording available. Are we free-threaded ready? Looking at where free-threaded Python fails
Free-threaded Python addresses the limitations of the Global Interpreter Lock (GIL), a mechanism that historically restricted CPython to executing one thread at a time per...
-
Recording available. Do you know how well your model is doing? Evaluate your LLMs
Evaluating Large Language Models (LLMs) is critical for benchmarking performance, ensuring safety through railguarding, and verifying that fine-tuning actually improves model...
-
Recording available. Beyond Stateless: Why Your Web Service Architecture is Fighting Against Performance
Traditional web service architectures often rely on a stateless model where business logic is separated from data. In this pattern, every request requires fetching the object...
-
Recording available. Solving Marketplace Cold Start at Scale with Ranking
Marketplace cold start occurs when new listings lack the historical data—such as clicks, bookings, and reviews—required by ranking models to achieve high visibility. This...
-
Recording available. Exploring Germany's Urban Geography with Census and OpenStreetMap Data
Urban geography examines the physical, social, and economic aspects of city life, specifically focusing on the equitable distribution of resources and infrastructure. A primary...
-
Recording available. Ship Data with Confidence: Declarative Validation for PySpark & Pandas
Data validation in PySpark and Pandas pipelines often suffers from silent failures, such as NaN errors, unexpected null values, or duplicate rows resulting from joins. While...
-
Recording available. Programming Quantum Networks in Python
Quantum networks connect quantum computers to enable applications impossible for classical systems, such as blind quantum computing—where a server processes data without...
-
Recording available. 5 Years of NiceGUI: What We Learned About Designing Pythonic UIs
NiceGUI is a Python-based web UI framework designed to simplify the creation of graphical interfaces for robotics, automation, and general-purpose applications. It addresses...
-
Recording available. A minimalist introduction to Ansible
Ansible is an open-source automation tool written in Python used for configuration management, application deployment, and infrastructure provisioning. It operates on a push...
-
Recording available. From Pixel to Payouts: A Multi-Agent System for Real-Time Insurance Claims Processing
Insurance claims processing for vehicle damage is traditionally a slow manual process, often taking weeks or months due to the sequential nature of investigation, cost...
-
Recording available. Surviving AI Fatigue: Staying Sane and Relevant in a Fast Moving Field
AI fatigue is the mental and emotional exhaustion resulting from the constant bombardment of new artificial intelligence tools, research, and paradigms. This state is...
-
Recording available. How to create effective data visualizations
Effective data visualization requires a clear, written message to guide the audience, often integrated directly into the chart title. A robust mental model for this process is...
-
Recording available. Open Table Formats in the Wild™ - Reloaded: Vortexing Ducks over Floating Icebergs
Apache Iceberg addresses the limitations of Hive and Parquet by providing a table abstraction over cloud object storage. While Hive relies on partition-level metadata, Iceberg...
-
Recording available. (Autism and) The Predictive Brain Theory (in Tech)
The Predictive Brain Theory posits that the human brain does not function as a simple input-process-output system, but rather as a predictive engine. Instead of processing...
-
Recording available. Black Hole Stars: An Astronomical Mystery (Mostly) Solved with NumPyro and JAX
The James Webb Space Telescope (JWST) has revealed a population of compact, high-redshift objects known as little red dots. These objects appear in the early universe,...
-
Recording available. Beyond Vibe-Coding: A Practitioner's Guide to Spec-Driven Development in AI Engineering
Spec-driven development addresses the failures of "vibe-coding," where AI-assisted software engineering relies on fragmented chat sessions. This approach prevents common issues...
-
Recording available. Dynamic Knowledge Graphs
Dynamic knowledge graphs address the fragmentation of information across disconnected knowledge sources, such as Jira tickets, Git repositories, and API documentation. While...
-
Recording available. From Research Models to SLAs: Operationalizing TSFMs with Python
Operationalizing Time Series Foundation Models (TSFMs) involves transitioning from research-based models to production-ready APIs that meet enterprise service level agreements....
-
Recording available. Is digital sovereignty a new buzzword in AI development?
Digital sovereignty in AI development addresses the risk of dependency on foreign technology providers, political instability, and restrictive legal frameworks. A primary...
-
Recording available. Restaurants around train stations are bad and I can prove it
This analysis investigates whether restaurants located near train stations in Germany are systematically lower in quality than those in city centers. The study utilizes a...
-
Recording available. Why Did The Model Do That? Debugging the Ghost in the Machine
Explainable machine learning (XAI) addresses the challenge of interpreting the decision-making processes of complex models, such as Convolutional Neural Networks (CNNs), Vision...
-
Recording available. Building reliable data pipelines with polars and dataframely
Data pipelines often suffer from reliability issues due to inconsistent data types, poor memory management, and silent runtime failures. To address these problems, a...
-
Don’t Let Imposter Syndrome Win: U Can Do Big Things from a Small Place, A 7-Year African AI Journey
The provided transcript contains no technical content, subject matter, or data regarding artificial intelligence. The text consists entirely of a moderator announcing the...
-
Recording available. How to Search Through 800 Billion Records in Real Time
Searching through a threat repository containing one trillion records requires a high-throughput pipeline capable of aggregating data from hundreds of microservices. The...
-
Recording available. Making Tech Tutorials Accessible: Practical Techniques for Educators
Technical tutorials often exclude users due to situational, temporary, or permanent impairments. Addressing these barriers requires a shift from the medical model of...
-
Recording available. Roll for Architecture: DungeonPy – A D&D Companion as Server + Thin Clients
DungeonPy is a distributed system designed to facilitate remote Dungeons & Dragons sessions by separating the game state from the user interface. The architecture employs an...
-
Recording available. Sentinel Values in Python: Semantics, Double Dispatch, and the Limits of Typing
Python's `None` keyword is frequently used to represent missing values, defaults, or errors, but this versatility creates ambiguity when `None` is also a valid value within a...
-
Recording available. Simulating the World using SimPy: A practical Example
Explainable AI (XAI) addresses the "black box" problem in machine learning, where models may rely on spurious correlations—such as identifying a wolf based on snow in the...
-
Recording available. When Space Weather Breaks Your GPS: Building an Explainable Early Warning System
Space weather, driven by solar activity such as flares and coronal mass ejections, disrupts the ionosphere and causes plasma density fluctuations known as Large-Scale Traveling...
-
Recording available. 7 Anti-Lessons from Building a PydanticAI Agent: Mistakes We Made So You Don't Have To
Building an AI agent for compliance intelligence in the life sciences industry requires balancing autonomy with strict regulatory oversight. Initial attempts using a...
-
Recording available. Accelerate FastAPI Development with OpenAPI Generator
In this presentation, Dr. Evelyne Groen, Senior MLOps Engineer at Malt, and Kateryna Budzyak, Senior Machine Learning Engineer at Malt, demonstrate how to accelerate FastAPI...
-
Recording available. Building MCP at the Speed of Hype: Principles That Outlast the Trends
Model Context Protocol (MCP) development requires a strategic approach to manage the rapid evolution of AI frameworks and LLM releases. The primary challenge is the "speed of...
-
Recording available. Destructive Testing: 10 Practical Ways to Expose Hidden Application Risks
Destructive testing identifies hidden application risks by intentionally attempting to break a fully deployed system. While developers typically focus on unit and integration...
-
Recording available. Escape the Hype: Teaching LLM Concepts Through an Interactive AI Factory Game
The AI literacy gap occurs when users adopt large language model (LLM) technologies without understanding their failure modes, cost implications, or latency constraints. To...
-
Recording available. Fight your garbage data: implementation of a pythonic data quality monitoring framework in PySpark
IoT data quality monitoring in PySpark addresses the challenge of "garbage in, garbage out," particularly when dealing with diverse software versions and unstable connectivity...
-
Recording available. From Ticket to Draft: How Munich Automates Citizen Inquiries with AI
The City of Munich is automating citizen inquiries for the driver's licensing authority to reduce the administrative burden of repetitive requests. The city processes over five...
-
Recording available. How to compare apples with oranges: Proper evaluation of article-level demand forecasts
Article-level demand forecasting at scale involves predicting sales for hundreds of thousands of items across multiple markets, resulting in approximately 10 billion records...
-
Recording available. Is my AI Recruiting biased? - How to evaluate these systems
AI recruiting systems typically function as black boxes where a Large Language Model (LLM) receives a job description and a candidate's CV via an API to produce a numerical...
-
Recording available. Letting AI Move: Robotics Demos Powered by Python
Robotics provides a tangible medium to demystify artificial intelligence, transforming it from an abstract black box into customizable software. By utilizing the Reachy Mini, a...
-
Recording available. Octopus AutoML: Extracting Signal from Small and High-Dimensional Data
Octopus AutoML is an open-source supervised machine learning library designed specifically for small, high-dimensional tabular datasets, such as those found in clinical trials...
-
Recording available. Pair & Share: How formal Mentoring pushed REWE Analytics to a new level
The REWE Analytics department, consisting of approximately 150 data scientists, machine learning engineers, and analysts, implemented a formal mentoring program called Pair &...
-
Recording available. Personalized Restaurant Recommendations at Scale combining Transformer with Gradient-Boosted Ranking
The Universal Venue Ranker (UVR) addresses the challenge of personalizing restaurant and retail recommendations at scale while balancing the tension between recurring user...
-
Recording available. Schema-Driven Lambdaliths in Python with AWS Lambda Powertools and Pydantic
Large-scale serverless applications often suffer from operational overhead when split into numerous small functions. In high-frequency data analysis environments, such as air...
-
Recording available. Small Language Models for Tool Calling Are Better Than You Think
Small Language Models (SLMs), typically defined as models with under 4 billion parameters, offer significant advantages over Large Language Models (LLMs) regarding latency,...
-
Recording available. Wetterdienst: Fast, Unified Access to Open Weather Data with Polars
In this presentation, Benjamin Gutzmann, a Data Engineer at Otto Group data.works, introduces Wetterdienst, a Python library designed to simplify the complex process of...
-
Recording available. When LLMs Are Too Big: Building Cost-Efficient High-Throughput ML Systems for E-Commerce Cataloging
E-commerce cataloging requires classifying millions of product offers into thousands of categories across multiple languages in real time. High-throughput systems must process...
-
Recording available. Zero-Copy or Zero-Speed? The hidden overhead of PySpark, Arrow & SynapseML for inference
Scaling machine learning inference to 6 billion daily predictions using an Ensemble LightGBM model requires overcoming the performance bottleneck created by the Python-JVM...
-
Recording available. Accuracy Is Overrated: Ship Stable Forecasts (Without Lying to Yourself)
Forecasting instability occurs when retraining a model on new data causes the long-term forecast to shift dramatically, even if the overall accuracy improves. This vertical...
-
Recording available. Building Trust in Your Data Pipelines with Observability
Data pipeline observability addresses the problem of "black box" pipelines, where engineers only discover failures through stakeholder complaints or cryptic error messages. To...
-
Recording available. Demystifying Containers with Python: Building a Minimal Engine from Scratch
Containers are isolated processes that share the host system kernel rather than running on a separate guest OS like virtual machines. This isolation is achieved through Linux...
-
Recording available. Empowering Data Scientists with Zero Platform Friction: Deploying Streamlit & Friends in 3 Minutes
In this presentation, Bernhard Schäfer, Senior Data Scientist at Merck, and Nicolas Renkamp, Global Head of Platform Product Portfolio at Merck, address the common "wall" data...
-
Recording available. Holistic Optimization: Implementing "Pipeline-as-a-Trial" HPO with Ray and Cloud Infra
Local optimization in machine learning occurs when a specific model is tuned for its own output rather than the performance of the entire downstream system. This often leads to...
-
Recording available. Tracking Knowledge Diversity in LLM-Generated Responses.
Measuring knowledge diversity in Large Language Model (LLM) responses addresses the problem of information redundancy, where models generate synonymous content across different...
-
Recording available. Before You Ship Your Agent: An Agent Builder’s Primer on Jailbreaking Attacks
Large Language Model (LLM) agents introduce significant security risks because they combine three dangerous capabilities: the processing of untrusted external input, access to...
-
Recording available. Build a web coding platform with Python, run in WebAssembly
Web-based coding platforms often face security risks when executing user-submitted code on a server and consistency issues when users rely on varying local Python...
-
Recording available. Don’t call your LLM too often! How to build your dialog graph with confidence and sleep at night.
Large Language Model (LLM) integration in corporate environments often leads to excessive operational costs and system inefficiencies due to redundant API calls and complex,...
-
Recording available. Embedding Data Science in IoT devices with MicroPython and emlearn
Embedding data science into IoT devices is achievable using MicroPython and specialized libraries to overcome the memory and processing constraints of microcontrollers. A...
-
Recording available. Foundation Models in Forecasting: Are We There Yet? Lessons from the Trenches
Zero-shot time series forecasting aims to predict future data without task-specific training, utilizing foundation models to handle multivariate settings and global forecasting...
-
Recording available. From Row-Wise to Columnar: Speeding Up PySpark UDFs with Arrow and Polars
PySpark User Defined Functions (UDFs) often introduce significant performance bottlenecks due to serialization overhead and row-by-row execution. Standard Spark UDFs rely on...
-
Recording available. How We Built an Inclusive Data Organization: Careers, Community & 50% Women
Building an inclusive data organization requires a systemic approach that addresses hiring biases, career progression, and cultural retention. A primary challenge in the...
-
Recording available. It Works on My Machine: Why LLM Apps Fail Users (Not Tests)
Large Language Model (LLM) applications often fail in production despite passing automated tests because of a gap between technical performance and user expectations. This...
-
Recording available. Mastering the Hex: A Case Study in Reinforcement Learning for Strategy Games
The project addresses the challenge of developing an autonomous AI agent to play a turn-based strategy game on a hexagonal grid. The game involves capturing territory, managing...
-
Recording available. Metashade: Compilerless Immediate-Mode Shader Generation in Pure Python
Metashade addresses the challenges of shader programming, specifically portability across different rendering APIs, the permutation explosion in real-time shading, and the lack...
-
Recording available. Post-Processing and Visualization of Astrophysical Data with PyPLUTO
PyPLUTO is a Python package designed for the post-processing and visualization of numerical astrophysical simulations generated by the PLUTO code. The primary challenge...
-
Recording available. Practical Refactoring with Syntax Trees
Automated refactoring involves transforming source code by representing it as data. While regular expressions suffice for simple changes, complex refactoring requires syntax...
-
Recording available. Process, Analyze, and Transform Python Code with ASTs
Abstract Syntax Trees (ASTs) represent the structure of Python source code as a tree of language constructs, such as modules, classes, and functions. In the Python compilation...
-
Recording available. Production ML across 2015-2035: A Journey to the Past and the Future
Production machine learning has evolved from early foundational efforts to a complex ecosystem of specialized operations. The initial "Genesis" period focused on the...
-
Recording available. Python Hates Being PID 1: Writing Container-Aware Code for Kubernetes
Python processes running as PID 1 in Kubernetes containers often fail to respond to SIGTERM signals because the Linux kernel treats PID 1 as a special init process that ignores...
-
Recording available. Reaching the next level of abstraction: meta classes and what they enable
Metaprogramming involves writing code that controls other code rather than manipulating data. In Python, this is possible because classes, functions, and instances are all...
-
Recording available. Simplicity Scales: Rewriting to a Django Monolith and Monorepo
A legacy system characterized by a fragmented microservice architecture, unclear ownership, and high infrastructure costs created a poor developer experience and slow release...
-
Recording available. State of In-Browser ML: WebAssembly, WebGPU, and the Modern Stack
In-browser machine learning leverages WebAssembly (Wasm) and WebGPU to execute code and models on the client side, reducing server costs and improving user privacy. WebAssembly...
-
Recording available. The Day the Agent Started Lying (Politely)
LLM agents often suffer from silent degradation, where the model continues to provide confident responses despite a shift in the underlying data distribution. This problem is...
-
Recording available. Tidy Finance in Practice: How Explicit Assumptions Avoid Bad Investment Strategies
Investment strategies often suffer from misleading performance metrics due to implicit assumptions and data manipulation. A common pitfall is the omission of extreme negative...
-
Recording available. Vibe NLP for Applied NLP
Applied Natural Language Processing (NLP) often faces a tension between LLM maximalism—using a single large model as the entire system—and traditional software engineering....
-
Recording available. What Breaks When Automatic Speech Recognition Systems Go Multilingual
Multilingual Automatic Speech Recognition (ASR) systems face significant challenges regarding linguistic variance, acoustic features, and data integrity. A primary problem is...
-
Recording available. AI Is Changing the Game: Building Modular, AI-Ready Platforms on Top of Legacy Systems
Large enterprises often struggle with expensive, tightly coupled legacy systems that hinder innovation due to rigid release cycles and high structural risk. In highly regulated...
-
Recording available. Simplifying RAG Document Pipelines with Multimodal Embeddings
Retrieval Augmented Generation (RAG) pipelines often struggle with PDF documents because these files are display-oriented rather than data-oriented. Traditional ingestion...
-
Recording available. Demystifying Parallel Programming in Python: from CPU to quantum processors, including GPU and TPU
Parallel programming in Python requires an understanding of hardware architecture to optimize performance. Central Processing Units (CPUs) rely on clock frequency, physical...
-
Recording available. Building Non-Biased Synthetic Datasets: What Actually Works (and What Fails)
Generating non-biased synthetic datasets for sensitive domains, such as the legal sector, addresses the challenge of strict privacy regulations like GDPR and the EU AI Act,...
-
Recording available. Hype, Hope, or Headache? Making Sense of GenAI, LLMs, and AI Agents with Anecdotal Evidence
The current trend toward Generative AI and Large Language Models (LLMs) creates a risk of neglecting traditional data science. Over-reliance on LLMs for simple problems...
-
Recording available. Architecture Under Constraints: Designing Systems That Still Evolve
Software architecture in regulated environments, such as medical device software, requires treating constraints not as downstream friction but as primary design inputs. The...
-
Recording available. AsyncIO vs Threads: who survives in the No-GIL Era?
Python concurrency for I/O-bound tasks traditionally involves a choice between multi-threading and AsyncIO. Multi-threading utilizes OS-level threads that are serialized by the...
-
Recording available. Securing AI Agentic Systems: Enforcing Safety Constraints in AI Agent
Securing AI agentic systems requires shifting from a reliance on natural language instructions to the implementation of rigorous software engineering constraints. A primary...
-
Recording available. Building Secure Environments for CLI Code Agents
Running CLI code agents directly on a host machine introduces significant security risks, including accidental data exfiltration, destructive file system operations, and the...
-
Recording available. "You are an intelligent business analyst": how i learned to talk to business
Technical professionals often struggle with the gap between development and business requirements, frequently viewing communication as a distraction from coding. This...
-
Recording available. Demystifying Agentic AI Using Small Language Models
Agentic AI relies on the ability of a language model to function as a decisioning engine. This process involves a reasoning loop where a prompt is decomposed into a plan, which...
-
Recording available. Free T(h)r(e)ading: A Trading Systems Journey Beyond the GIL
Algorithmic trading systems are ideal for testing Python's free-threading capabilities because they are latency-sensitive, combine I/O and CPU-bound tasks, and require strict...
-
Recording available. Heat: scaling the Python scientific stack to HPC systems
HEAT is an open-source Python library designed to scale scientific data analysis from local workstations to high-performance computing (HPC) systems. It addresses the memory...
-
Recording available. Come for the Code, Stay for the People.
The Python community is built on a foundation of inclusivity and accessibility, evolving from a niche group of enthusiasts into a global ecosystem. While the technical...
-
Recording available. Designing and Scaling a Python Library in the Open: Architecture, Automation and Community
ScanAPI is a Python-based CLI framework designed to test APIs by executing tests defined in YAML specifications. To address the challenges of scaling an open-source...
-
Recording available. How to mix conda and pip without causing “environmental” damage.
Python packaging involves two distinct ecosystems: the pip ecosystem and the conda ecosystem. Pip, managed by the Python Packaging Authority (PyPA), utilizes the Python...
-
Recording available. PyTorch and CPU-GPU Synchronizations
PyTorch executes GPU operations asynchronously, allowing the CPU to schedule tasks and run ahead of the GPU. Performance degradation occurs during CPU-GPU synchronization,...
-
Recording available. Your First Open Source Contribution in Python: From Fork to Pull Request
Open source software consists of public code that is accessible, modifiable, and distributable under specific licenses. These licenses define the obligations of contributors,...
-
Recording available. Agent-Based Hyperparameter Optimization for Gradient Boosted Trees
Hyperparameter optimization for Gradient Boosted Trees, such as LightGBM and XGBoost, typically requires tuning 14 to 19 parameters. While frameworks like Optuna use Bayesian...
-
Recording available. Array-Oriented Programming in Python: Libraries, Techniques, and Trade-offs
Array-oriented programming in Python shifts the focus from individual element manipulation to operations on entire data structures. This paradigm addresses the performance...
-
Recording available. Beyond Kafka and S3: Python Data Pipelines with HTTP-Native Bytestreams
Cross-organizational data sharing often incurs high alignment costs due to differing infrastructure, technology stacks, and compliance regimes. Traditional solutions like...
-
Recording available. Getting Career Clarity in Uncertain Times
Career clarity in volatile professional environments requires a shift from traditional, linear career models toward intentional self-reflection and experimental planning. The...
-
Recording available. Learnings Building DevOps as a Software Engineer
Establishing a DevOps culture in a non-software manufacturing environment requires a phased transition from zero infrastructure to a hardened, automated pipeline. The initial...
-
Recording available. Making bad CLIs fun with Small Language Models
Complex command-line interfaces (CLIs) often suffer from steep learning curves due to intricate syntax and obscure flags. While large language models (LLMs) can translate...
-
Recording available. No, you can't 'eval' your way to fairness
Algorithmic fairness cannot be achieved solely through quantitative metrics or evaluation libraries because fairness is a lived experience rather than a calculable state. The...
-
Recording available. On Interventional Generalisation
Interventional generalization addresses the challenge of determining whether a specific action will improve a desired outcome in novel, unseen situations. Standard machine...
-
Recording available. Scaling Data Processing for Training Workloads at DeepL Research with Rust
Training large-scale AI models requires processing terabytes of data across distributed workers. A proprietary file format using GZIP blocks and JSON dictionaries previously...
-
Recording available. SQL is Dead, Long Live SQL: Engineering reliable analytics agent from scratch
Reliable analytics agents can be engineered by combining an in-process analytical database with a structured agentic loop and semantic context. Using DuckDB as the core engine...
-
Recording available. To nest, or not to nest? Nested data types in Polars with big data
Polars provides nested data types, specifically lists and arrays, to store multiple values per row. While lists allow variable lengths, arrays require fixed lengths. These...
-
Recording available. Ty mypy: The New Generation of Python Type Checking
Ty is a high-performance Python type checker developed by Astral, designed to address the performance bottlenecks and ergonomic issues associated with MyPy. While MyPy remains...
-
Recording available. Your Data Is Leaking: A Hands-On Introduction to Differential Privacy with OpenDP
Differential privacy (DP) addresses the failure of traditional data anonymization techniques, such as k-anonymity, cell suppression, and the removal of personally identifiable...
-
Recording available. "Honey, I vibe coded some crypto" - Security in the age of LLMS Keynote
Large Language Models (LLMs) introduce significant security risks when used for "vibe coding," where developers rely on AI to generate complex logic, such as cryptographic...
-
Recording available. The Multimodal Era of Machine Learning (and How Python Made It Possible) Keynote
Multimodal machine learning integrates diverse data types, such as text, audio, images, and video, to enable representation learning, cross-modal retrieval, zero-shot...
-
Recording available. A View of Sovereignty from The Cloud Keynote
Digital sovereignty is the capacity to act independently regarding computing infrastructure. While cloud computing is often abstracted as a virtual service, it relies on...
-
Recording available. From Scratch to Scale: Turning LLM Code into Architecture Insights Keynote
Modern large language model (LLM) development faces a significant challenge in the lack of detailed technical reports, as many open-weight models are released with minimal...
-
Recording available. From Prompt to Production: How to use AI Code Assistants for Python Data Systems
Building production-ready Python data systems with AI coding agents requires moving beyond linear prompting toward a structured framework of skills, personas, and state...
-
Recording available. The Art of the Optimal: A Pythonic Approach to Complex Decision-Making
Complex decision-making problems, such as optimizing a car assembly line's paint shop, are often addressed using heuristics or greedy algorithms. In a scenario where vehicles...
-
Recording available. Python in Climate Tech: Vehicle-to-Grid
Vehicle-to-Grid (V2G) technology addresses the volatility of renewable energy by utilizing electric vehicle (EV) batteries as distributed energy storage. In Germany, the EV...
-
Recording available. Kickstart Coding at Scale: How Project Template Automation Unlocks Developer Productivity
Project template automation streamlines the bootstrapping and maintenance of software repositories at scale. By using Copier instead of Cookiecutter, developers can generate...
-
Recording available. Making my Apache Spark™ talk more interesting using AI
Apache Spark serves as a versatile transform layer for large-scale extract, load, and transform (ETL) pipelines. It is particularly effective for processing massive datasets,...
-
Recording available. From Hard Problems to Proven Solutions: Solving Decision Problems with Gurobi
Mathematical optimization addresses complex decision problems by systematically finding the best possible solution within a defined set of rules. Every optimization problem...
-
Recording available. Panel: Evolution, Revolution, or Illusion? The Future of Python and Coding in the Age of AI
The integration of Large Language Models (LLMs) into software development has shifted the role of the programmer from manual syntax implementation to high-level architectural...
-
Recording available. Leveraging Hexagonal Architecture When Building Applications
Hexagonal architecture, also known as ports and adapters, is a software design pattern used to isolate core business logic from external technologies such as databases and...
-
Recording available. The foundation model revolution for tabular data
Tabular prediction is shifting from bespoke models toward foundation models, addressing the historical failure of deep learning to consistently outperform gradient boosting...
-
Recording available. Type Errors for Better Agent-Assisted Development
AI coding agents often struggle with validation, frequently missing call sites during refactors or introducing runtime errors by ignoring optional types. While linting is fast...
-
Recording available. Start-Ups & Investors
Founding a startup requires a transition from technical research to operational management, often necessitating a shift in focus from product development to sales and...
-
Recording available. Closing Session
Open source software engineering requires a shift in narrative to better communicate the value that professional engineering standards and open source culture provide to...
-
Recording available. Lightning Talks 1
The session focuses on community announcements, professional development, and technical challenges within the Python ecosystem. A series of event announcements highlight...
-
Recording available. Lightning Talks 2
Community events and technical lightning talks focused on Python applications, data science, and professional development. Key announcements included PyCon Africa 2026 in...
-
Recording available. Open Source as a Business — Models, Paths, and Practice
Building a business around open source requires distinguishing the software's distribution method from the commercial business model. Open source serves as a community and...
-
Recording available. Opening Session
PyCon DE and PyData focus on the intersection of open-source software and artificial intelligence, emphasizing a culture of cross-pollination and practical implementation over...
-
Recording available. Panel What Do We Still Need to Learn?
The integration of AI agents into the professional workforce creates a tension between cost-reduction strategies and the need for long-term capability development. Many...
-
PyLadies Fireside Chat
The integration of Artificial Intelligence (AI) and Large Language Models (LLMs) is fundamentally altering the Python ecosystem, shifting the role of the developer from a...
-
Recording available. Stop Waiting, Start Shipping: Real-World Strategy for Open-Source LLMs
The deployment of open-source Large Language Models (LLMs) focuses on moving beyond the search for a single "savior" model toward a strategy of specialized harnesses and...
-
Innovation Day: Startup Lounge [no-video]
Presented by hessian.AI, the Startup Lounge at PyCon DE & PyData 2026 introduces a dynamic, unrecorded format designed to bridge the gap between open source, AI, and...
-
Problem Clinic: Python in Regulated Environments --- What Works, What Doesn't [no-video]
In this session from PyCon DE & PyData 2026, Alexander CS Hendorf, an independent AI and open-source strategy advisor, leads a critical examination of the challenges facing...
-
Workshop: What do we still need to learn? [no-video]
In this interactive workshop, Paula Gonzalez Avalos, Sebastian Neubauer, a Sr Staff Data Scientist at Blue Yonder, and Dr. Kristian Rother lead a community-driven exploration...