A conversation on the Agents and Engineers podcast about treating coding agents as collaborators: Marimo Pair, notebooks that outlive an agent's context window, and reproducible agentic data science.

A SciPy 2026 talk on Canvas Chat, a browser-based tool for non-linear, visual AI-assisted data analysis with Python and LLMs.

Pair-programming with a coding agent inside a Marimo notebook to build a Bayesian hierarchical model of protein melting points, live.

A hands-on Festival of Genomics 2026 workshop on coding agents inside Marimo notebooks, with live protein data analysis and 3D structure visualization.

Marimo as the analysis canvas and Cursor as the coding agent: live notebook EDA and 3D protein visualization.

Battle-tested patterns for AI coding agents: plan, write tests first, let the agent build, document, and repeat.

A live demo with Hugo Bowne-Anderson: Pixi lock files, multi-environment workflows, and CUDA-enabled Docker containers for reproducible data science.

Testing your code, your data, and your pipelines: why software testing matters for open source and data science.

Standardize, test, document, and version: software skills that turn data science from solo work into teamwork.

From UBC iGEM to MIT to industry: how network science, deep learning, and Bayesian methods show up in biological engineering.

Graphs as arrays, message passing, and how they add up to message passing neural networks, explained with lots of pictures and a minimum number of equations.

Five principles borrowed from software development that accelerate data science projects and amplify their impact.

How data scientists can incorporate testing into their workflow, with Hypothesis for code and Great Expectations for data.

My PyCon 2019 rant against canned statistical procedures, and how Bayesian estimation with PyMC3 extends A/B testing beyond two groups.

In which I show that Bayesian deep learning is nothing special: priors on the weights, explained with pictures and PyMC3 code.

Turn "which statistical test do I use?" into Bayesian estimation problems, with PyMC3 as the workhorse.

Networks are everywhere, and representing problems as graphs unlocks ones that are otherwise hard to reason about.