PyData London 2025
2025-06-06..2025-06-08
London, United Kingdom
[PyData London 2016, PyData London 2017, PyData London 2018, PyData London 2019, PyData London 2022, PyData London 2024, PyData London 2025, PyData London 2026]
| title | author | length |
|---|---|---|
| Bayesian Time Series | Chris Fonnesbeck | 01h 27m |
Practical checklist: tutorial uses PyMC in Jupyter (Conda / Pixie setups) to build AR/AR1/AR2, random‑walk and state‑space models in PyMC, add Fourier… | ||
| Cutting Edge Football Analytics in Python | Joris Bekkers | 32m |
Shows how Polars Keras Spektral Kloppy and unravel sports process 7 million frames across 64 matches into one-graph-per-frame inputs and build an expected… | ||
| Hands-on with Apache Iceberg | Anders Bogsnes | 01h 12m |
Interactive, hands‑on walkthrough that actually writes and reads Iceberg tables: end-to-end example using PyIceberg with a LakeKeeper catalog and MinIO/S3… | ||
| Transformers Inside Out (Parts 1 & 2) | Sam Joseph, Ph.D | 01h 26m |
A runnable mini‑GPT gets built with nanoGPT and PyTorch, trained on a ~35,000 knock‑knock corpus (max sequence 15, vocab ≈10,000) using a custom prepare.py and… | ||
| Package Your Python Code as a CLI | Jeroen Janssens | 01h 26m |
Shows exactly how to turn a Python script into a CLI: copy code, pick arguments, read from stdin/stdout, add a shebang, chmod +x and put the file on PATH, then… | ||
| Graph Theory for Multi-Agent Integration: Clinical Use Cases | Ahmad Albarqawi | 01h 04m |
Claims a concrete pipeline: an open‑source IntelliNode graph library (Apache‑2 2023) orchestrates speech, vision, recommender and LLM agents, with… | ||
| Predicting Weather Patterns using Time Series ML | Suyash Joshi | 01h 16m |
Hands-on claim: convert Open-Meteo hourly 2024 CSV into line protocol and ingest into InfluxDB processing engine with a Python VM, install neuralprophet on… | ||
| GPU Accelerated Python w: CUDA | Jacob Tomlinson | 02h 55m |
Claims practical demos: a transparent pandas plugin that dispatches to cuDF and a scikit‑learn plugin, hands‑on use of Numba CUDA kernels and CuPy within the… | ||
| Workshop on developing Reinforcement Learning | Ade Idowu | 01h 31m |
Workshop walks through concrete code and outcomes: Q‑learning SARSA DQN implemented from scratch, then gymnasium environments (grid world, FrozenLake… | ||
| Keynote- Tony Mears | PyData London 25 | Tony Mears | 38m |
Title Innovation is Dead claims innovation is on life support but salvageable and gives concrete evidence and fixes: global genomics costs collapse to cost of… | ||
| Quantum Computing with Python | Andrea Melloncelli | 01h 27m |
Hands-on Python notebooks demonstrate building circuits, creating Bell and GHZ entangled states, and running them on a simulator and on IBM cloud hardware via… | ||
| Building Vertical Agent with AG2 AgentOS | Chi Wang, Tim Santos | 01h 27m |
Open source AG2 AgentOS claims that composing small specialized agents is often faster and cheaper than one big model and demonstrates that with working demos… | ||
| Debugging Leadership | Matthew Upson | 30m |
Claims six concrete leadership pitfalls framed as Python errors including ValueError: self-worth not defined and DeadlockError: unable to release control… | ||
| How To Measure And Mitigate Unfair Bias in Machine Learning Models | John Sandall | 01h 31m |
Hands-on workshop that builds a biased hiring pipeline and then fixes it: generates 400 synthetic CVs with OpenAI, injects names and clues, trains sklearn /… | ||
| LIGHTNING TALKS | 45m | |
Lightning talks run five minutes each and hit concrete takeaways: GPU Python pathway advice with CuPy and RAPIDS and reusable CUDA kernels plus an Accelerated… | ||
| PyScript - Python in the Browser | Chris Laffra | 39m |
PyScript runs Python in the browser by bootstrapping Pyodide or MicroPython compiled to WASM and loading packages on demand. Demos include a Python-driven… | ||
| Conquering PDFs- document understanding beyond plain text | Ines Montani | 29m |
Claims that PDFs are a bad source of truth and that extracting a unified structured format first makes document pipelines manageable. Uses Dockling plus a… | ||
| How we unified feature engineering across data and backend | Alex Jones | 23m |
Monzo describes a system that defines features once and reuses them offline and online using BigQuery, Spark (batch and streaming on EMR), a key-value store… | ||
| From Next Token Prediction to Reasoning and Beyond | Jay Alammar | 45m |
Claims concrete improvements: models move from next-token training to multi-step reasoning by combining SFT then RL for reasoning with tool-enabled rollouts. A… | ||
| LLM Inference Arithmetics | Luca Baggi | 34m |
Explains how parameter counts map to VRAM and compute so numbers mean something: Gemma 2B is closer to a 3B as a parameter-vs-storage example, a 7B model… | ||
| NetworkX is Fast Now- Zero Code Change Acceleration | Mridul Seth | 36m |
Claim: zero code change acceleration lets NetworkX dispatch algorithms to backends like GraphBLAS and cuGraph via an environment variable and a plugin system… | ||
| Multi-tasking fraud detection From Trees to MLPs | Callum Court | 24m |
Monzo trains multi‑task neural nets in PyTorch alongside LightGBM/XGBoost baselines and finds statistically significant uplift in recall, with the best… | ||
| Media Mix Modelling | Natalia Ziemba Jankowska | 29m |
Bayesian Media Mix Modelling shows how attribution, adstock and spillover, and daily-level channel data pin down which buys actually move revenue. A deployed… | ||
| Parallel PyTorch Inference with Python Free-Threading | Michał Szołucha | 33m |
A concrete claim: building and running a Python 3.13-T free-threaded build lets multiple Python threads run PyTorch inference truly in parallel. The demo… | ||
| AI agents testing How to evaluate the unpredictable | Emeli Dral | 37m |
Projects prioritize unit, integration and end‑to‑end tests wired into CI/CD. Prompts and tools are treated as functions; validate formats with JSON schema… | ||
| Why you should stop pretending your sparse data is dense | Alex Owens | 28m |
Compares concrete formats and measurements: pandas (NumPy-backed) uses NaN/NAT/None and promotes int to float, Arrow keeps the full data buffer plus an Arrow… | ||
| Keep Calm and Data On | Leanne Fitzpatrick | 45m |
Crisp, measurable claims: data‑center emissions hit about 3% of global emissions (2024 estimate), US private AI investment tops $109B in 2024 versus $9.3B… | ||
| Enhancing Fraud Detection | Radion Bikmukhamedov | 28m |
Company profiles are auto-generated per section using an SOP as the prompt and take about a minute or two to produce, then encoded as embeddings or tokenized… | ||
| Building a Real-World Adverse Media Pipeline | Adam Hill | 39m |
Comply Advantage describes a deployed adverse‑media pipeline that processes 1 million articles a day, uses LLM APIs rather than hosted models, and balances a… | ||
| Platforms for valuable AI Products- Iteration, iteration, iteration | John Carney | 27m |
Argues that rapid, repeated experiments are the path to value, with concrete tradeoffs and tooling choices: use Google MLOps stage two for a reference… | ||
| Bringing stories to life with AI | Olena Kutsenko | 37m |
Claim: generative agents actually evolve in real time by wiring LLMs into streaming infrastructure. Concrete stack: Apache Kafka + Flink pipeline streams 20–30… | ||
| Data Challenges for Scaling Multi-Agent GenAI Apps with Python | Theo van Kraay | 42m |
Practical takeaways and concrete trade-offs: use LangGraph or Semantic Kernel but expect them to mis-handle mixed operational+vector workloads; embed vectors… | ||
| Successful Projects through a bit of Rebellion | Ian Ozsvald | 38m |
Claims a private leadership community called Rebel AI peer mentorship group grew in 15 months and yields concrete tactics not fluff, shows a few multi‑million… | ||
| Sovereign Data for AI with Python | Lex Avstreikh | 24m |
Claims a full Pythonic sovereign AI stack that runs on air gapped or hybrid infrastructure and uses open source pieces: PyPI and Conda library server, a… | ||
| How Narwhals brings Polars, DuckDB, PyArrow, & pandas together | Marco Edward Gorelli | 36m |
Narwhals is a lightweight compatibility layer that makes code accept and return Polars DuckDB PyArrow pandas and cuDF without forcing conversion to pandas… | ||
| Diving into Transformer Model Internals | Matt Squire | 34m |
Walks through Hugging Face code: how AutoTokenizer and AutoModel resolve huge mapping dictionaries to concrete classes, how model files include a config and… | ||
| CUDA in Python- A New Era for GPU Acceleration | Andy Terrel | 35m |
Concrete claims and tools: Python bindings for CUDA (nvmath-python, cccl-python, cuda-core, numba-cuda) plus cuPy, cuDF and cuPy‑numeric/Legate stack that… | ||
| Transfer Learning: Leveraging Pretrained Models with Limited Data | Salman Khan | 13m |
Claims that transfer learning finds laughter in very small labeled sets using pretrained audio encoders and simple classifiers; dataset is ~5s clips with… | ||
| Making LLMs reliable - A practical framework | Lena Shakurova | 38m |
A practical framework appears: define metrics and tests up front, run experiment monitor improve cycles, add real‑time input and output guardrails, and use… | ||
| Control all of your projects with copier! | Tim Paine | 30m |
Copier claims to let one central template repo control many projects by prompting with a copier.yml, rendering Jinja, and then later running copyor update to… | ||
| Scaling AI workloads with Ray & Airflow | Tatiana Al Chueyr | 29m |
astro-provider-ray (Apache-2) claims to let Airflow create and delete Ray clusters on Kubernetes, submit Ray jobs and monitor progress in the Airflow UI… | ||
| Is coding assistant as good as we thought | Cheuk Ting Ho | 29m |
Claims that coding assistants boost productivity and test coverage, but evidence is specific: a homebuilt agent (a failed precursor) runs a deepseek LLaMA file… | ||
| Analysing smart meter data to uncover energy consumption patterns | Sofia Pinto | 29m |
Uses a representative sample of 13,000 households with half‑hourly gas and electricity from 2019, engineers features (proportion overnight, seasonal curves)… | ||
| Building a knowledge graph for climate policy | Harrison Pim, Fred O'Loughlin | 38m |
An open-source knowledge graph combines in-house policy expertise with scalable data infrastructure to label 27.5k documents (2M pages, 45M passages) and 350k… | ||
| Journey Towards Deep Learning for Ranking | Mihail Douhaniaris, Theodore Meynard | 30m |
GetYourGuide replaces an XGBoost ranker with a PyTorch transformer and deploys it within nine months while keeping high throughput and low latency. The… | ||
| You Came to a Python Conference. Now, Go Do a PR Review! | Samiul Huque | 29m |
Covers concrete PR-review actions: identifies three bad-review archetypes and prioritizes technical bugs vs nitpicks; walks through Python pitfalls with… | ||
| Agentic Cyber Defense with External Threat Intelligence | Jyoti Yadav | 27m |
An autonomous agentic pipeline ingests WHOIS, GeoIP and VirusTotal feeds, extracts deterministic features (domain age, country, thread hits) and LLM-derived… | ||
| Building Inclusive Machine Learning Models | Elizabeth Osanyinro | 35m |
Case studies name concrete harms: the Apple credit card shows a 20x gap in limits, a health‑care algorithm routes people of color to $1,800 less spending on… | ||
| Automating Porosity Detection with Deep Learning | Onyekachukwu Ojumah | 24m |
Claims that CNN segmentation models plus EfficientNet transfer learning convert X‑ray CT stacks into pixelwise porosity maps, spotting pores missed by human… | ||
| Reproducibility in Embedding Benchmarks | Isaac Chung | 32m |
Shows concrete fixes and numbers: MTEB (and MMTEB/MIEB) adds task and model metadata, stratified subsampling and embedding caching to cut workload (tasks down… | ||
| nothing matches | ||