Varun Bhadauria — Software Engineer

I am currently part of Meta’s AI Infra team, focusing on scaling training and inference for recommendation system (RecSys) workloads using LLMs and long user sequences. My work spans optimizing state-of-the-art training and inference stacks built on PyTorch 2, exploring modern GPU architectures, and developing custom kernels in Triton to push the boundaries of performance and efficiency. Previously, I designed and implemented large-scale schedulers for LLM and RecSys training jobs, orchestrating workloads across 100,000+ GPU clusters. Earlier in my career, I worked on messaging infrastructure for WhatsApp, where I scaled tier-1 services to support over a billion monthly active users. My foundational experience includes developing file systems, block I/O kernel drivers, and SSD caching software for enterprise systems. I later transitioned to consumer gaming, contributing to the PlayStation 5 compression engine and building modern game authoring systems leveraging advanced I/O compression and deduplication technologies.

Stackforce AI infers this person is a highly skilled engineer in AI infrastructure and embedded systems.

Location: Lafayette, California, United States

Experience: 20 yrs 2 mos