Sreedhar

I'm Sreedhar, an AI Architect with over 12 years of experience building production infrastructure for machine learning systems. I specialize in GPU-accelerated inference pipelines, LLM serving architecture, and the cost optimization problems that every ML team faces when moving from prototype to production.

My career spans the full lifecycle of ML infrastructure — from designing multi-node GPU clusters with InfiniBand interconnects to tuning vLLM PagedAttention kernels for maximum throughput. I've deployed inference systems serving millions of tokens per day across providers like NVIDIA H100, H200, and AMD MI300X clusters.

OpenGPU Radar started as an internal tool — a way to track cloud GPU spot prices and calculate VRAM requirements without relying on vendor-provided estimates. When I realized how fragmented and opaque the cloud GPU market was, I decided to make the data public. Every price point, every benchmark, and every VRAM calculation on this site is independently sourced and reproducible.

My technical focus areas include LLM inference optimization (PagedAttention, speculative decoding, quantization-aware serving), multi-node GPU cluster architecture with NVLink and InfiniBand, and deterministic cost modeling for long-context workloads. I built the VRAM calculator and the entity graph that powers the model, GPU, and provider pages on this platform.