Raghu Nallamothu

WORK HISTORY

EDUCATION

SKILLS

ABOUT RAGHU NALLAMOTHU

I\'ve spent 15 years building the infrastructure layer not managing it from a distance, but getting into the hard problems and staying there.At Oracle I spent six years building two of the most foundational layers of Oracle Cloud: the compute infrastructure (think EC2) and the block storage service (think EBS). When I joined, these were early-stage systems on small teams. By the time I left, we were at thousands of nodes, exabytes of data, 99.995% uptime. I architected features, dealt with distributed systems problems that only show up at real scale. Those six years are where I developed my real understanding of what cloud infrastructure requires the hardware constraints, the failure modes, the operational discipline that doesn\'t show up in any architecture diagram.From Oracle I went to Facebook, where the problem was different but the stakes were the same. I built data infrastructure systems processing trillions of events a day designed the architecture, wrote it in Python, and delivered several 0 to 1 products that ended up embedded in Facebook\'s global infrastructure. What that period gave me was a ground-level understanding of network and data infrastructure at a scale most systems never reach, and what it takes to build something that doesn\'t get rewritten six months later.At Ambient I joined as a Senior Engineering Manager and was promoted to Director. I now run the full platform engineering org: AI/ML infrastructure, data infrastructure, and compute and hardware platforms. We\'ve cut infrastructure costs by more than 50% and held 99.95% uptime while launching a managed SaaS platform that has brought in our largest enterprise clients.On the AI infra side, we\'re building the systems that make vLLMs usable in production - inferencing infrastructure, LLM gateways, Prompt registries, and the platform that research scientists and product teams need to build and run AI applications reliably. Most of the problems aren\'t model problems they\'re infrastructure problems that look like model problems until you dig in.On the data infra side, the challenge is rebuilding data infrastructure for what AI actually needs. The pipelines, storage patterns, and processing systems that work for analytics don\'t work for vLLM training and inference workloads. Compute and hardware sit underneath all of it. Having built both the compute and storage layers of a cloud provider, I have a different relationship with hardware constraints and that perspective changes how I think about everything above it.