ML PLATFORM ENGINEERING
Deep Learning Training Platform
Architected a unified ML training platform to schedule workloads across multiple Kubernetes GPU clusters and led a team of approximately 10 engineers, collaborating with product managers and ML researchers.
01 /Platform architecture
Architected an internally developed, end-to-end machine learning training platform that sits in front of the Kubernetes APIs and provides unified scheduling across multiple GPU clusters. The platform orchestrates training workloads across clusters, while each cluster schedules training pods onto its GPU worker nodes.
02 /Technical leadership
Led a team of approximately 10 engineers, driving cross-functional collaboration with product managers and machine learning researchers.
03 /Automated annotation
Developed models for automated data annotation, accelerating model training and deployment.
04 /Training performance
Performed performance analysis on machine learning training jobs and resolved bottlenecks to improve training efficiency.