Systems Engineer
I work on AI inference infrastructure at Google, where I've built and led teams as a TL/M. Currently, my work includes modelling efficiency and performance optimizations towards improving SDLC at Google. I have built and shipped Context Caching products on Vertex AI and Google AI Studio from preview to GA.
Before that, I created ghOSt, a userspace scheduling framework that moves Linux scheduling decisions out of the kernel and into userspace, enabling rapid policy iteration without kernel recompilation. ghOSt is deployed on Google Search, Google Cloud and published at SOSP 2021. I drove cross-org efforts to de-fragment Google's compute fleet across Cloud and Shared Borg, designing the technical framework to mint new tiers of capacity from resources that were previously stranded.
Earlier work included tracing persistent Cloud VM jitter and fleetwide networking stalls back through the stack to root causes in the Linux kernel's memory subsystem and building comprehensive solutions: fragmentation-resistant pinning, SLAB cache reaper latency improvements (10ms → 50μs), and bespoke memory allocators (SNAP SOSP 2019).
Before Google: Qualcomm Innovation Center (ARM64 server enablement, upstream Linux kernel). Startups (Embedded Systems). MS Computer Science, Penn State.
Kernel and userspace components for delegating Linux scheduling to userspace, enabling rapid iteration on scheduling policies without kernel recompilation.