Rohit D.

Every AI model you've used this week ran on a GPU cluster that someone had to build, qualify, and bring online at scale. That's what I do. I'm a hardware systems engineer at Meta, where I own the end-to-end quality arc for AI accelerator platforms — from first build at manufacturing partners through mass production and sustaining one of the world's largest GPU fleets. In plain terms: I'm accountable for making sure the hardware that trains frontier AI models actually works, at yield, on schedule, every program. Over the last decade across Meta, AWS, and Silicon Labs, I've: → Built an ML-powered debug system that cut hardware fault diagnosis from 2 days to 4 hours across 25,000+ production records — saving 100+ engineer-hours per week for a 50-person team → Stood up US-based GPU HW System manufacturing capability under a tariff-driven deadline with no domestic precedent — delivered $100M+ in savings without missing schedule → Established liquid cooling and optical interconnect integration at an ODM from zero — a capability that didn't exist before, delivered at <0.5% defect rate for next-generation high-TDP AI clusters → Maintained quality accountability across a 150,000+ GPU fleet — the infrastructure that powers Meta's AI research and products What makes my work unusual is the combination: end-to-end NPI ownership from prototype to mass production, the ability to operate directly on ODM factory floors and in vendor QA organizations simultaneously, and AI-native execution — I've built and deployed LLM-powered engineering workflows that change how teams operate, not just how I work.