I owned two generations of machine learning training platforms at Reddit.

I inherited and owned our Kubeflow Pipelines based platform and scaled it from a peak sustained utilization of 50 nodes and 2,000 CPU cores to 250 nodes and 10,000 CPU cores to sunset.

Thereafter I owned our Ray based training platform and scaled it from zero to a peak sustained utilization of 1,000 nodes and 40,000 CPU cores.

I was a full-stack (but primarily backend) engineer at Spell, a machine learning training and inference platform startup. Acquired by Reddit.