Writing
Long-form notes on AI systems, engineering leadership, design systems, and building in public — written the way I think aloud, not the way a press release sounds.
GOMAXPROCS, CPU Limits, and the Kubernetes Trap That Silently Kills Go Throughput
If your Go pod has less than one full core, the runtime treats it as single-threaded for parallelism. For years the Go scheduler didn't know about Kubernetes CPU limits and over-scheduled anyway. Here's what changed in Go 1.25, why giving a pod 'extra' CPU sometimes dramatically improves throughput, and how to get it right in 2026.
Expert Vision: Cognitive Foundations for Human-AI Collaboration
What fifteen years of studying how experts perceive, decide, and perform under pressure — from eye-tracking systems to competitive gaming to enterprise AI platforms — teaches us about building agents that genuinely extend expertise rather than merely automate tasks.
Reliability Engineering for Generative AI Platforms
Fifteen years of distributed systems, real-time pipelines, and incident command applied to LLM platforms. How to build agentic systems that degrade gracefully, contain failures, and remain auditable when everything is on fire at 2 a.m.
Context Engineering for Production Agentic Systems
Beyond prompt engineering lies context engineering — the systematic design of memory, state, retrieval contracts, and compression layers that turn brittle LLM workflows into reliable, observable, enterprise-grade agentic platforms.
What My Side Projects Are Really About
ChillStep Radio, quantum chaos, synesthetic pianos, and a robots.txt for AI — the side projects I build when enterprise platforms go quiet and curiosity takes the wheel.
The Boring Go Layer Between Your RAG Demo and Production
RAG demos live in notebooks. Production RAG lives in ingestion pipelines — and I keep choosing Go for the unglamorous middle.