San Francisco, California, United States
TLM of TPU Fungibility team. My team designed, launched, and maintained two products central to LLM model development lifecycle on Tensor Processing Unit (TPU) 1.) Virtualization layer that enables dynamic migration of ML workloads (pretraining, distillation, finetuning, inference) across heterogeneous accelerators 2.) Intelligent Load Balancer for LLM inference
- Co-founded the company and successfully pitched and raised our Seed and Series A funding from Greylock and Sequoia Capital. - Acquired by OpenAI for $500M
Deal Sourcing: Leveraging my engineering background from a major Greylock portfolio win (Arista), I sourced and evaluated early-stage startups across cloud, DevOps, and infrastructure. Greybase: To track key investment signals, I needed a way to organize fragmented data scattered across inboxes, Crunchbase, LinkedIn, and 50+ other silos. I built an internal tool, Greybase, to ingest and query all of it in one place. Initially for my own use, this grew to be the firmwide OS.
I joined Arista as one of the first 40 employees, a formative "ground floor" experience that allowed me to build the foundational distributed systems for a company that eventually redefined the networking industry. Core Distributed Systems: Developed the crash-persistent event logging and remote authentication mechanisms used by Arista’s earliest enterprise customers. Resilient Infrastructure: Re-engineered critical sections of the Spanning Tree Protocol (STP) to ensure high availability during the company’s pre-IPO hyper-growth. Stack: C++, Python, and TACC for agent-based systems.