A new season for data & ML.

The Miami Marlins’ Research & Development organization supports player evaluation, forecasting, and data-driven baseball operations. As the department expanded its analytical capabilities, it needed an engineering foundation that could keep up.

Together, the Marlins and CloudKite built that foundation: scalable cloud infrastructure, an automated ML workflow, and engineering practices that gave the team more time for its core work.

The challenge

Before partnering with CloudKite, the Marlins depended on on-premises infrastructure. Network hardware, application servers, and databases ran on physical servers in their offices. Their engineers handled maintenance, deployments, access control, and support manually.

The operational overhead reached well beyond the infrastructure itself:

• Engineers spent as much as 40% of their week on support and maintenance.
• Existing systems could not scale with new data sources and ML initiatives.
• There was no unified cloud environment.
• Support work made it difficult for people to specialize.
• New hires took significant time to gain access and become productive.

The organization wanted to modernize so analysts and ML engineers could experiment faster, improve forecasting, and build the tools that support baseball decisions.

The partnership

The Marlins chose CloudKite for its Google Cloud experience, existing MLB relationships, and commitment to modern engineering practices.

We embedded with their team to support cloud adoption, teach engineering fundamentals, and co-design workflows. The work went beyond implementing infrastructure: it helped establish the standards and working practices of a growing engineering organization.

CloudKite helped us go from zero to one. You taught us best practices, helped us design our cloud architecture, and worked with us to mature as an engineering organization.

— Brian Chase, Vice President of Baseball Systems, Marlins

The solution

Establishing a cloud-first architecture

The engagement began with discovery and planning alongside the Marlins’ engineering leadership and technical staff. We reviewed the on-premises environment, its performance bottlenecks, and the overhead required to maintain it.

Together, we mapped the goals of a migration to Google Cloud: better performance, more capacity for analytics, greater operational agility, cost optimization, and a resilient foundation that could grow with the organization.

Building the ML and data foundation

Entering the 2025 season, the Marlins did not yet have the shared infrastructure and standardized processes needed for scalable ML operations. Moving an experimental model into production required a more consistent approach.

CloudKite and the Marlins built an end-to-end MLOps foundation, with standards for data governance, model development, source repositories, and deployment practices.

Airflow and Argo CD ran on Kubernetes, integrated with GitHub and single sign-on through Google Workspace. The resulting workflows covered data ingestion, model training, continuous integration and delivery, and automated monitoring for model performance drift and data quality.

We provisioned the cloud foundation with Terraform and established separate development, staging, and production environments. The platform provided the compute and storage required for data processing and model training, together with a repeatable path into production.

Establishing engineering practices

The platform was supported by changes to how the team worked:

• Infrastructure as code for repeatable, reviewable changes.
• DevOps practices and automated delivery workflows.
• Standardized development and deployment environments.
• Clear operational practices and comprehensive observability.

These practices helped turn separate, ad-hoc initiatives into a more consistent ML operations capability.

Results and impact

Routine maintenance dropped from nearly half of an engineer’s week to less than ten percent. With standardized workflows and a cloud foundation in place, the team could spend more time building, forecasting, and exploring new ideas.

ML models in production

50+

Deployed during the 2025 season.

Reduction in maintenance time

87%

Engineers regained time for analysis, forecasting, and internal tools.

or less to onboard

30 min

Standardized access, least privilege, and automated reviews.

More time for the work that matters

Reducing maintenance gave analysts and engineers more room to specialize in baseball analysis, forecasting, and internal tools. Engineers regained nearly a full workday that had previously gone to routine support.

If we’re spending 40% of our time on maintenance, we’re not being as impactful to putting the best team on the field. CloudKite changed that.

Faster, safer onboarding

Standardized roles and access workflows replaced a slower, manual process. Least-privilege access, multi-factor authentication, and automated reviews helped make onboarding predictable while reducing access risk.

The Marlins team reported that adding a new employee took less than 30 minutes. With the new IAM structure and standardized permissions, new hires could become productive on their first day.

A path from experiment to production

More than 50 ML models were deployed in 2025. The automated pipeline made it possible to operationalize those models during the season and created a more consistent path from an idea to a working production system.

The result was faster experimentation, more reliable forecasting, and more capacity to build the internal tools the organization needed.

A stronger engineering organization

The partnership helped the R&D organization establish clearer ownership and more consistent ways of working:

• Engineering practices: shared standards and clear responsibility for systems.
• Scalable infrastructure: a cloud and DevOps foundation that could grow with the team.
• Organizational clarity: documented workflows and repeatable processes.

With less time absorbed by maintenance, specialists had more capacity to focus on their own disciplines and collaborate on new capabilities.

Looking ahead

The Marlins’ modernized engineering foundation supports faster experimentation, stronger forecasting, and a scalable data ecosystem. The infrastructure, workflows, and practices built together give the organization a platform for continued development as its needs evolve.

Technologies used

• Google Cloud
• Google Kubernetes Engine
• Terraform
• Argo CD
• Apache Airflow
• GitHub
• Google Workspace
• Tailscale
• Helm
• Python
• R
• Posit