Gojek scales ML operations and cuts costs with Flyte

Industry

Logistics

Use Cases

Data Processing
Model Training
Inference

Challenge

Gojek needed a scalable, Kubernetes-native orchestration platform.

Gojek is one of Southeast Asia’s largest multi-service and digital payment platforms, offering 20+ services across mobility, food, commerce, and financial services. In 2021, the Gojek ecosystem contributed 1.7% of Indonesia’s GDP, making operational reliability and scalability critical.

Gojek’s machine learning systems power complex use cases such as:

  • Pickup point generation
  • Driver–rider matching
  • Voucher and incentive allocation
  • Pricing optimization

Beginning in 2017, ML pipelines were built with an abstraction layer on top of Clockwork and Airflow, now hosting more than 1,000+ DAGs. While the abstraction enabled containerized execution, it introduced major problems:

  • Difficult local development — pipelines couldn’t be easily tested outside containers
  • Steep learning curve for data scientists
  • Scheduler scaling issues, missed DAGs
  • Significant infrastructure overhead

The system would not sustain Gojek’s growth.

The platform “probably wouldn’t get us through the next five years.” — Pradithya Aria Pura, Data Science Platform

Gojek needed a scalable, data-aware, Kubernetes-native orchestrator with an intuitive developer experience.

“Flyte is fast and scalable… it provided us with good results.”

Pradithya Aria Pura

Principal Software Engineer at Gojek

Solution

Flyte delivered scalability, strong versioning, and a superior developer workflow.

After evaluating multiple orchestrators, Gojek chose Flyte, the engine behind Union Cloud.

Key capabilities that drove the decision:

1. Workflow versioning

Mission-critical for rolling back to previous versions when needed.

“There are only a few platforms that provide this kind of versioning… it’s critical.”

2. Kubernetes-native execution

Flyte runs naturally on Kubernetes, eliminating infrastructure strain and improving reliability.

3. Powerful Python SDK (flytekit)

Dynamic workflows, flow control, and rich abstractions improved developer productivity.

“Flytekit provides a powerful and expressive way of building pipelines.”

Gojek began migrating legacy ML pipelines and onboarding teams to Flyte.

100
-200

new workflows/month expected on Flyte

20
-80%

cost savings across migrated pipelines

1000
+

legacy DAGs on path to deprecation

Results

Flyte enables rapid workflow growth and significant cost savings.

Gojek will fully deprecate its legacy ML pipeline platform within six months. Meanwhile, the company is rapidly expanding Flyte adoption, expecting 100–200 new workflows each month.

Migrated workflows already show substantial operational efficiency:

“From several pipelines we migrated to Flyte… we see around 20% to 80% cost saving.” — Pura

Flyte provides the scalability, reliability, and developer experience needed to support Gojek’s high-growth, multi-service ecosystem—positioning the company for the next five years and beyond.