Anaconda Scale Customer Event
Case Study

How Mozilla Scaled Machine Learning Without Moving Its Data

Mozilla unified its machine learning stacks on AI orchestration, now part of Anaconda, and kept sensitive data in its own cloud.

COMPANY SIZE
501-1000
INDUSTRY
Technology
LOCATION
San Francisco, CA
FOUNDED
2005
One Platform

Replaced separate machine learning stacks across teams

Sensitive Data Stays Put

Workflows run inside Mozilla’s own Google Cloud project

More Frequent Deployments

Teams that struggled to run jobs on a schedule now run them regularly

Mozilla has always put data privacy and user protection first, and that focus shapes how it collects data, builds machine learning (ML) models, and automates its work. Chelsea Troy, a lead engineer at Mozilla, has been central to the effort to streamline its ML operations and scale them efficiently. This case study shows how Mozilla moved from fragmented, disjointed ML infrastructure to a streamlined, scalable system on Metaflow-based AI orchestration.

The Challenge: Separate Stacks, Separate Problems

Before adopting AI orchestration, Mozilla’s ML teams operated independently, each with its own stack and deployment tools. Chelsea put it this way: “Each team was using wildly different stacks that largely amounted to whatever it is they knew at the time they deployed their model.”

The cost showed up in two places. Deploying and managing models was hard, and some teams couldn’t even run their jobs on a regular schedule. That held back Mozilla’s ability to scale ML across teams.

Troubleshooting was the second problem. Logs and monitoring were decentralized, so tracking an issue across teams took time. “There was a refactor that made a fix universal, and it broke that job, and no amount of debugging seemed to be able to figure it out,” Chelsea shared.

The stakes were rising. Mozilla wanted to bring ML models into consumer-facing products like Firefox translations, which meant scaling deployment while protecting data privacy. “We have a small team at Mozilla, so duplication of effort is particularly costly for us,” Chelsea explained. The organization needed a solution that increased efficiency, reduced duplication, and centralized operations.

Why Managed Metaflow?

When Chelsea and her team went looking for a unified platform, they evaluated open-source Metaflow, MLflow, and other orchestration tools. Mozilla’s requirements were strict, especially on privacy. As a privacy-first company, it couldn’t use solutions that stored data outside its own infrastructure, and its sensitive ML workflows had to stay inside its Google Cloud Platform (GCP) project. “It was extremely important for those machine learning flows that ran on sensitive data to remain inside our GCP project,” Chelsea noted.

The managed platform built on Metaflow by Outerbounds stood out. Metaflow is the open-source framework created at Netflix, where it supports 3,000+ AI and ML projects. Anaconda acquired Outerbounds in April 2026, and that platform is now AI orchestration in the Anaconda Platform. Anaconda continues to contribute to Metaflow as an open-source project.

AI orchestration let Mozilla run ML jobs inside its own GCP environment, so sensitive data stayed put. It also gave multiple teams one place to deploy, monitor, and scale, without the overhead of separate stacks.

The Metaflow API was a natural fit for Mozilla’s data scientists, too. “The Metaflow API integrates so seamlessly with the Python code that teams are already writing. It’s a light lift,” Chelsea explained. Data scientists could adopt AI orchestration without overhauling their existing workflows, which was a key factor in the decision.

The Results

More Frequent Deployments

One of the first changes Mozilla saw was the ability to run ML jobs more often. Teams that had struggled to deploy consistently could now do so with ease. “Many teams that didn’t run their jobs regularly are now able to do so through the platform,” Chelsea noted.

With deployment infrastructure off their minds, teams could focus on improving model quality and running more experiments. Removing the manual barriers to running jobs let Mozilla put its ML capabilities to work across the organization.

Faster Model Integration

Integrating models had been slow, largely because each team managed its own infrastructure and deployment tools. “Each team was managing its own deployment stack, which often meant that deploying models could be slow and cumbersome,” Chelsea explained.

After the move to AI orchestration, integration became far more streamlined. “Once they saw how easy it was to integrate their models, they were like, ‘I can do this with my model by next week,’” Chelsea recalled. Teams could now deploy faster, spend less time on infrastructure, and spend more time optimizing their models. That let Mozilla experiment and iterate more often.

Clearer Logs and Error Tracking

Monitoring was a long-standing pain point. Before AI orchestration, troubleshooting production issues meant manual intervention, and logs were hard to reach, so engineers spent significant time debugging.

Centralized logs and error tracking changed that. “The ability to access logs easily and see what’s happening saved us countless hours of troubleshooting,” Chelsea explained. Engineers can now diagnose and fix issues quickly and respond to production problems faster.

What Made the Difference

Chelsea pointed to four reasons AI orchestration worked for Mozilla:

  • Runs in Mozilla’s cloud: AI orchestration works inside Mozilla’s GCP environment, which met its privacy requirements.
  • One platform: Mozilla brought its ML models onto a single platform, simplifying operations across teams.
  • Easy to adopt: The Metaflow-based API was simple for data scientists to pick up, which kept onboarding light.
  • Responsive support: The team behind AI orchestration, now part of Anaconda, responded quickly to feature requests and built solutions that fit Mozilla’s needs.

Together, these let Mozilla focus on building better ML models, with no overhead from separate stacks and manual processes.

Looking Forward

Mozilla is expanding its ML work across more projects, including further development of Firefox’s translation models and other consumer-facing applications. With AI orchestration as its core ML platform, it has the tools to scale that work while holding to its commitment to data privacy and transparency.

“We’ve freed up engineering time and resources. Instead of worrying about infrastructure, we can now focus on building better machine learning models,” Chelsea concluded.

The result is a stronger foundation for future innovation, built on the principles that define Mozilla: privacy and user protection.

Want to scale AI from development to production on your own terms? Contact our team for a personalized demo.

“We’ve freed up engineering time and resources. Instead of worrying about infrastructure, we can now focus on building better machine learning models.”

Chelsea Troy
Lead Engineer, Mozilla

AI Orchestration

Discover how the Anaconda Platform can transform your data science workflows, while maintaining enterprise-grade security and governance.