Voice, standards, and the operating systems that hold them — from long-form narrative to launch messaging, built for technical companies with skeptical readers.
Writer, editor, and content strategist with 19 years across editorial publishing, agencies, startups, and the enterprise — I care about the craft of words and the systems that protect it.
From immunodiagnostics and crypto to data storage and cross-border transactions — the subject changes; the job doesn’t.
Not a wish list — the work I've actually done, in-house and for agency clients, from Fortune 500 programs to blogs and sites launched from nothing.
An LLM wiki, built with Claude Code — a git repo of markdown files the agent reads, writes, and maintains, where knowledge gets compiled once instead of re-derived every time you ask. Obsidian sits on top as the reader: wiki-link navigation and a graph view that shows whether the thing is actually connecting. I heard the pattern described in a session led by Rashim Mogha, Global Head of AI Innovation and Training Ops at NVIDIA and recognized it as an editorial problem wearing engineering clothes.
Mine covers the craft layer: voice construction, structural patterns, interview technique, post-mortems on pieces that worked and pieces that didn't. The premise is that a brief shouldn't start from zero.
Push back. If a source is weak, or a pattern I claim isn't supported by the pieces in sources, say so. A wiki that agrees with everything I put into it is worth nothing.
The pattern proves itself when the wiki answers something that isn’t in any single source, because the answer lives in the relationships between pages. The failure mode is the opposite: pages that summarize instead of concluding — note-taking wearing a wiki’s clothes.
Most people using AI for writing craft a better prompt. This builds the substrate the model works from — closer to what an editor actually does.