A year of mentor coaching PD lived in Canvas pages where nobody outside the program could see the design. No Rise license, so I couldn't build a course. I designed one instead. The deliverable is a one-page design doc (audience, objectives, alignment, assessment) and a build-ready spec, and the alignment table exposed what my own knowledge checks never verified.

A post-mortem of my KOReader patch for Entity Footnotes, the "no" that reshaped it, and why solving it right meant not touching the plugin at all.

A post-mortem on a self-hosted PWA that turns a photographed receipt into a split YNAB transaction using AI vision, and why the fix for drifting grocery categories turned out to be the same fix behind our program data rebuild.

Two years of overlapping spreadsheets drifted out of sync until nobody could agree on our own candidate count. How a three-dashboard architecture and AI-assisted reconciliation made the data trustworthy again, and the two rules that keep it that way.

A post-mortem on a self-hosted app that turns my Hardcover reading history into individualized, AI-generated book recommendations, verified against real catalog data, and why the build kept echoing the individualized-pathway work I do at NXT GEN Teach.

Most apprenticeship progress reports are generic by default and less than useful by consequence. How we rebuilt ours around the actual licensure standards for two different special education tracks, and why one form for two programs was the original mistake.

On Justin Reich's "Failure to Disrupt": why lasting change in education comes from the slow work of communities, not from the next disruptor or killer app.

What MOOC completion rates and the edtech Matthew effect reveal about who online learning actually serves, and who it leaves behind.

A brief look at Ed Tech as an accessibility tool, from the 18th-century hornbook to AI tutors, and why the tools were never the point.