I found inefficiencies in my own workflow. So I built tools to eliminate them.

Content updates used to begin with a writer opening a live support article, selecting text section by section, and pasting it into a working document, losing structure along the way and introducing copy errors under deadline pressure.

The extractor reads the article's live HTML directly, walks the DOM to preserve heading hierarchy, lists, and links, and writes a clean structured copy into a Google Doc ready for editing. What was a 20-minute error-prone chore per article became a few seconds of automation, and the working copy is now guaranteed to match the live source.

Editorial review does not scale: one editor's judgment, applied one document at a time. This tool encodes that judgment into a Google Docs sidebar that reviews as you write.

It scans the document against editorial style guidelines in real time, flagging vague directional words, context-dependent phrasing, and style-guide violations, and suggests concrete alternatives inline. The writer sees the correction while the sentence is still warm, which teaches the guideline instead of just enforcing it. The editor's standards are now present in every document, whether the editor is or not.

Country selector work meant gathering content from hundreds of market-specific article variants before any writing could begin: a full manual phase of opening, reading, and copying, multiplied by every market in scope.

The crawler walks the country selector systematically, requests each market's article, extracts the structured content, and assembles everything into a single edit-ready Google Doc organized by market. The gathering phase disappeared from the schedule entirely, and the coverage is exhaustive by construction: the script cannot skip a market the way a tired human can.

Team intake ran through two systems that did not talk to each other: work arrived in Buganizer, planning lived in Google Sheets, and a rotating human kept them in sync by hand. When leads were absent, the sync degraded and the pipeline with it.

Culvert is a bi-directional sync engine between the two. New Buganizer issues appear in the sheet with their metadata; sheet-side triage decisions flow back as Buganizer updates. During lead absences it ran the entire intake operation autonomously with zero pipeline disruptions. It was built independently, on the observation that the job needed doing, not on a ticket asking for it.

The CMS ingested enterprise XML; the writers were fastest in Markdown. The gap between the two was bridged manually, one reformatting pass per deployment, on every single content update.

This pipeline converts Markdown source files into the CMS's XML schema automatically: headings, lists, tables, cross-references, and metadata all mapped by rule rather than by hand. The manual reformatting step vanished from every deployment, cutting project turnaround by 20% and removing an entire category of formatting defects at the same time.