I build and operate production infrastructure on Kubernetes and bare metal. My deep domain is high-availability blockchain RPC, and I go all the way down into the engines: execution clients, LLM inference servers, and the systems bugs that hide in both.
A stale hardware gate locked NVIDIA's GB10 (DGX Spark) out of SGLang's fast sparse-decode kernel and crashed the server on startup. The fix is one line.
Read the writeup →A cancellation race on a shared EVM made eth_estimateGas non-deterministic. A leftover goroutine, cancelling the wrong request, quietly turned a failure into a success.
10 years in production infrastructure. Specialized in operating high-availability blockchain RPC platforms across Kubernetes and bare metal, with end-to-end ownership from GitOps and automation through observability and reliability.