Moving People and Products with Machine Learning on Kubeflow
summary
GO-JEK builds its driver allocation system on Kubernetes and Kubeflow at scale — supporting ~2 million drivers and ~100 million bookings per month — claiming measurable business uplift from ML served in-cluster. They deploy a feature store Feast (BigQuery + BigTable/Redis serving) to eliminate training/serving skew and duplicate feature work, use Fairing to turn notebooks into Docker jobs, and orchestrate repeatable runs with Kubeflow Pipelines, plus Terraform/GitOps for multi-region rollout.
AI-written summary · about
chapters (11)
- 0:00Start
- 0:19Takeaway Message
- 4:51Finding the right driver
- 8:37ML on K&s: Economies of scale
- 11:10ML on K&s: Feature Store (Feast)
- 13:07ML on Kos: Rapid expansion
- 19:24Using Feast and Kubeflow to build and deploy models
- 32:44ML Development Workflow
- 33:23Core Tenets
- 33:59System Diagram
- 34:26OSS Momentum!
related talks
51mMachine Learning using Kubeflow and Kubernetes
14mKubeflow: Where ML Meets Kubernetes
54mScaling Machine Learning Pipelines in Cloud
30mKubeflow, Your ML Tools on Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes
39mDeploy your ML Workloads on Kubernetes with Kubeflow
10mKeynote: Machine Learning using Kubeflow and Kubernetes
46mAccelerating Machine Learning DevOps with Kubeflow