Netflix Metaflow UI: Bringing clarity to data-science workflows

Netflix is one of the world’s most data-driven companies. From choosing what shows to produce to deciding what to recommend to each viewer, thousands of experiments and models run behind the scenes. To support this, Netflix created Metaflow, an open-source toolkit that helps data scientists manage their workflows.

While Metaflow’s command-line tools were powerful, keeping track of large-scale workflows quickly became difficult. Data scientists needed an easier way to see what was happening in real time, to spot errors faster, and to share the right information with colleagues. That’s where Codemate came in.

The challenge at Netflix

At Netflix, a single experiment might include hundreds or thousands of runs. Monitoring them through the command line alone made it hard to keep an overview and even harder to dig into details when something failed.

The Metaflow team wanted a graphical user interface that would be fast, intuitive and informative, without slowing down the underlying system. The goal was to improve the day-to-day experience for data scientists while supporting Netflix-scale production workloads.

Designing the Metaflow user interface (UI)

Codemate worked hand-in-hand with Netflix’s engineers to design and build the first Metaflow UI. The interface gives data scientists a clear visual view of their workflows, updated almost in real time.

Netflix had defined a few high-level goals for the project, which we used as guiding principles:

• The UI must answer a simple but critical question: “What is happening or has happened when ML models are run?”
• It should feel snappy, even with a run containing 100,000+ tasks and an archive of over one million runs.
• It needed to be easy to adopt and easy to deploy.

These goals shaped every design and development decision. The result is a UI where users can follow workflows visually, spot issues quickly, and share exact states with colleagues — without slowing down the system.

The UI was released as open source, so the global Metaflow community can benefit from it and extend it further.

Impact at Netflix

The Metaflow UI has quickly become part of daily life for Netflix’s data scientists. It makes it easier to monitor experiments, troubleshoot quickly, and collaborate with colleagues. As Brett Rose, Manager of ML infrastructure from Netflix, put it:

The tool is now running at Netflix scale, supporting more than a million workflow runs and proving its value in production. And the story doesn’t stop there — the UI has been released as open source so the wider data-science community can benefit too.

Watch the Metaflow UI in action

As Codemate was the main partner for designing and developing the Metaflow UI, Netflix decided to organise the release event together with us. In the recording below:

• Ville Tuulos (Outerbounds) — one of Metaflow’s original creators at Netflix and now CEO of Outerbounds, the company developing Metaflow further — shares practical tips on how to get started and how to extend the UI through plugins.
• Brett Rose (Netflix) demonstrates the Metaflow UI in action.
• Teemu Kemppainen (Codemate) explains the user flow and design process.

Watch the video to see the Metaflow UI live, and hear directly from Netflix, Codemate and Outerbounds.

FAQ

What is Metaflow UI?

An open-source graphical interface for monitoring and debugging Metaflow flows and tasks in near real time.

Does it handle production runs?

Yes. Netflix uses it at production scale; the UI is architected to avoid impacting running jobs.

Can it integrate with our logging/monitoring?

Yes—via plugins and external links (e.g., Kubernetes dashboards, CloudWatch/DataDog).

What is Outerbounds?

Outerbounds is a company founded by former Netflix engineers, including Ville Tuulos, who created Metaflow at Netflix. Today, Outerbounds develops Metaflow further and works with partners like Codemate to support the community and extend the ecosystem.