20+ years of experience as an Infrastructure & Data Center Engineer. Working on AI Infrastructure & Performance Engineering.
ai infrastructure
Companies burn millions on Snowflake/Databricks while Parquet+S3 delivers 95% of the value at fraction of the cost. Netflix mastered telemetry by focusing on what matters. Your data goldmine isn't in petabytes—it's in the kilobytes driving your core business. Measure what matters, not everything.
ai inference
A 10ms latency difference affects 50M users. A 5% inefficiency costs $3M monthly. Yet infrastructure decisions are made blind. Atlas transforms $60M/week in cloud chaos into visual intelligence, exposing $11M in waste and why us-east-1 dominance is fear, not optimization.
ai infrastructure
When 2000 GPUs spike from idle to 2MW in perfect sync for distributed training, 8ms reaction time means grid failure. VESTA predicts these spikes in <100μs by catching kernel-level precursors before they hit the bus. Built this after a decade battling harmonics. Prayer isn't going to cut it.
ai inference
We built the internet for 99% cache hit rates, but AI inference can't be cached. We centralized compute in 4 mega-hubs, but AI needs edge processing. The result? Your perfectly trained model takes 1.2 seconds to respond instead of 200ms. Here's why AI feels slow—and what it'll take to fix it.