Garrett Cobb

Platform · Learning

Course Engine

A reusable platform that turns a course's source material into a complete interactive study site, with lessons, cheat sheets, glossary, search and four kinds of quiz question. Every course is checked by automated quality gates.

The business problem

Course material arrives as scattered readings and knowledge checks. I wanted a repeatable process that turns any course into a complete study site, and that proves every learning objective is covered instead of assuming it.

Requirements I set

  • Any new course is a content task, not an engineering task.
  • Every learning objective is mapped, taught and tested, and the platform can show that.
  • Quizzes build retention by reviewing earlier material, not only the current module.
  • Works offline and degrades gracefully when browser storage is blocked.

Solution design decisions

  • Static-site architecture: Astro with MDX content, full-text search and diagrams, built ahead of time so a course is just files to host. A scaffold command creates each new course from a template.
  • Content model: four question types (single, multi, match, order), each tagged with its learning objective, difficulty and source.
  • Rules enforced by a validator: every answer option must explain why it's right or wrong, and from Module 2 on each quiz must draw 20 to 30% of its questions from earlier modules.
  • Learner progress backup and restore, with merge or replace.

How the project was run

  • Four-phase playbook: ingest and map (with a review gate), instructional design, build, verify.
  • Each course has an outline, gap list, coverage map, alignment matrix and fact-check log.
  • A release requires zero validator, type and link errors, a passing headless smoke test locally and on the deployed site, all 421 coverage items taught, and a 20/20 sampled fact-check.

Results

MeasureResult
First course96 pages and 248 quiz questions
Coverage421 of 421 coverage items taught, each traced to a line of the source material
Fact check20 of 20 sampled claims verified
Release gateZero validator, type and link errors; headless smoke test passing locally and on the deployed site

What I learned

Write the coverage map first. Once every objective has an ID, "is this course complete?" becomes a query instead of a judgment call.

How AI was used

What AI did: Claude Code co-wrote every commit, working from my content model and playbook.

What I did: I designed the platform, the content model, the validator rules and the four phase playbook, and I own the review gate at each phase.

How quality was controlled

  • Validator enforces question rules and spaced review across modules
  • Coverage map proves every objective is taught and tested
  • Sampled fact check before release

Deployed on Cloudflare Pages behind Cloudflare Access, so the course site itself is private. One course built so far: E054, Foundations of Project Management.

Predecessor (v1): Certification Study Platform

Mar 2026 · The Course Engine replaces this earlier platform, which I built first.

Free, self-paced study guides with timed practice exams for seven cloud, DevOps and security certifications: HashiCorp Terraform, Microsoft AZ-900, four GitHub certifications and CompTIA SecAI+.

  • Template architecture: each exam is a content file on a shared quiz engine, so seven modules shared one codebase.
  • Domain-tagged questions with explanations, best-score and attempt history, and per-domain breakdowns.
  • Hosted as lightweight services behind Nginx and Cloudflare.
  • All questions are original, with trademark disclaimers; no copied exam content.