I founded and built Burla, an open-source cluster-computing framework for Python, as its sole engineer. I shipped it as both a managed cloud platform and a self-hostable product.
At Burla, I designed a cluster-compute architecture with a single-function API, no master node, and a custom distributed message queue. It runs Python functions across 1,000+ VMs in under a second end to end.
I also built automatic environment replication and realtime log streaming for containers. My “Dynamic RAM” algorithm automatically scales workloads vertically to prevent OOMs.
At Day Zero Diagnostics, I co-led a redesign of genomic ML training infrastructure, scaling training to thousands of CPUs and cutting runtime from two weeks to six hours. I also designed a custom data lake for 200+ TB of genomic data; earlier, at Blustream Corporation, I applied Q-learning to reduce app notification damage events by 36%.
