Skip to Content
View all events

marimo for AI and ML Development

Published by O'Reilly Media, Inc.

Intermediate content levelIntermediate

Enable reactive execution and predictable AI workflows

What you’ll learn—and how you can apply it

  • Articulate why reactive, dependency-aware execution prevents the reproducibility failures endemic to notebooks like Jupyter
  • Build and share reproducible AI/ML experiments using modern environment and dependency tooling alongside marimo’s controlled execution model
  • Develop interactive data exploration and model evaluation workflows with reactive visualizations
  • Apply AI coding agents as active development partners to accelerate prototyping, debug models, and iterate faster across the full ML workflow
  • Package and deploy marimo notebooks as standalone scripts, shareable web apps, or importable Python modules, without rewriting code

Course description

Interactive programming environments are central to modern AI development, but traditional ones like Jupyter notebooks don’t quite fit the bill. Join Parul Pandey to get a hands-on introduction to marimo, the next-generation programming environment designed specifically for AI and ML.

Learn how to do serious computational work in a reproducible environment that also brings data to life in new ways and allows practitioners to rapidly answer difficult questions about data and models—substantially decreasing time to value. You’ll also get a look at other modern tools that every AI and ML developer should adopt, as well as best practices for using AI coding agents for AI/ML development.

This live event is for you because...

  • You’re an AI or ML engineer or data practitioner who writes Python and builds models.
  • You work with data and run experiments, notebooks, and ML workflows in production settings.
  • You’re curious about marimo and want to see how it improves on traditional notebooks.

Prerequisites

  • A computer with marimo installed, or work in marimo’s free cloud-hosted notebook workspace Molab
  • Intermediate Python proficiency
  • Basic familiarity with data science or ML workflows (loading data, training a model, evaluating results, etc.)
  • Experience with Jupyter notebooks or similar interactive environments

Course preparation:

Recommended follow-up:

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Why interactive programming environments matter for AI and ML (45 minutes)

  • Presentation: Importance of interactive environments in the modern AI/ML stack; execution order and state in notebooks; how marimo enables reactive execution and predictable workflows
  • Group discussion: Issues that undermine correctness, reproducibility, and iteration speed
  • Hands-on exercise: Predictable execution and connected data and computation
  • Q&A

Reproducibility as a baseline for trustworthy AI (40 minutes)

  • Presentation: The “It works on my machine” problem; why reproducibility must be built into the environment itself
  • Hands-on exercise (guided in marimo): Identifying hidden dependency and environment assumptions in an existing experiment; examining how libraries, system settings, and execution context affect results; making dependencies explicit and controlled; reviewing how the same work behaves locally, in an editor, and in a different environment
  • Q&A
  • Break

Why interactivity accelerates AI discovery (35 minutes)

  • Presentation: Interactive computation as a unified system; data, models, visualizations, and user input working together in a single system that updates automatically
  • Hands-on exercise (guided in marimo): Building an interactive computation where data, model outputs, and visualizations update automatically; exploring and transforming data interactively and feeding results directly into models without breaking flow; adding interactive visualizations to examine intermediate results; using visual feedback to guide small experimental changes in real time
  • Group discussion: Where this interactive workflow differs from a traditional notebook approach
  • Q&A

How to use AI coding agents for AI/ML development (35 minutes)

  • Presentation: AI coding agents integrated directly into your development environment; when to trust AI assistance and where it provides value in AI/ML workflows
  • Hands-on exercise (guided in marimo): When to let AI agents write code; generating, modifying, and extending code within an active workflow; providing execution context, including data structures, model outputs, and dependencies; observing how context affects the quality of AI assistance; evaluating local versus cloud-hosted setups based on cost, speed, privacy, and control
  • Q&A
  • Break

From interactive work to reusable systems (25 minutes)

  • Presentation: Moving beyond private experimentation to create work others can run, inspect, and build upon; turning your work into executable scripts; running interactive code from the command line, passing arguments, and integrating into larger ML pipelines
  • Hands-on exercise (guided in marimo): Turning interactive code into an executable script and running it from the command line; passing arguments; integrating into a larger ML pipeline; publishing the same workflow as an interactive data app; exporting outputs to standard formats as published artifacts; importing functions and classes from the interactive notebook into a larger code base as proper Python modules
  • Q&A

Your Instructor

  • Parul Pandey

    Parul Pandey is an author and a data scientist who focuses on making machine learning clear, useful, and safe. She coauthored the O’Reilly book Machine Learning for High-Risk Applications and was the first woman in India—and the second worldwide—to be named Kaggle Grandmaster in the notebooks category. She has worked at H2O.ai as a principal data scientist and at Weights and Biases as a machine learning engineer and speaks at events, mentors community groups, and writes about her work in the data science space.

    linkedinXlinksearch

Skills covered

  • AI & ML
  • Design Patterns
  • Prompt Engineering
  • GPT

Sponsored by

  • marimo logo