Technical writing on React architecture, LLM engineering, and the tools I build with. All posts are static — fast to read, easy to cite.
"Predict the output" questions trip up almost everyone — not because the event loop is hard, but because nobody taught the three buckets. A foolproof mental model for synchronous code, microtasks, and macrotasks.
A build-it-this-week playbook for agentic engineering — deploy Claude Code as a PR reviewer on GitHub in under an hour with the official action, tune it into a teammate, roll it out on a Day-1/Week-1/Month-1 plan, and lock in the guardrails and metrics that keep it safe.
A hands-on, fully local ComfyUI pipeline that pairs the Next Scene LoRA (text-to-screenshot) with first-last frame video generation — turning 'prompt and pray' video into predictable, director-grade scene transitions.
Claude Code skills are packaged instructions that make Claude repeatable. A practical, step-by-step walkthrough of authoring one — using a code-review skill as the worked example — from folder structure to invocation.
RSC isn't a rendering optimization — it's an architecture primitive. A field-notes take on the server/client boundary, what belongs where, and why it mirrors the integration seams I build at work.
A deep dive into ComfyUI's node graph architecture — building reproducible AI image pipelines, caching strategies, and optimization patterns for production-grade synthesis.
A practical framework for prompt architecture — structured templates, few-shot patterns, context windows, and deploying local open-weights models with reproducible outputs.