A large library of licensed courseware sat on a hard drive as Word documents — genuinely good material, licensed for commercial resale, and completely unusable as a product in that form.
The obvious route was a hosted course platform: upload, set a price, pay a percentage of every sale forever. The arithmetic on that gets uncomfortable at volume, and it means the catalogue lives inside somebody else's business.
So we built the platform: catalogue, checkout, member accounts, course player, progress tracking, certificates. Then — the part that actually mattered — we built the pipeline that turns the source manuals into finished courses.
That pipeline is why there are fifty courses and around 1,600 lessons rather than the three or four that hand-building would have produced. Each course arrives complete, with its workbook, its narrated video and its certificate, from one run.
- A full learning platform — catalogue, checkout, member accounts, course player, progress tracking and certificates, on our own stack rather than a hosted course product
- An extraction pipeline that reads the original manuals and produces properly structured multi-session courses — 50 of them, at full depth rather than summaries
- A branded learner manual per course, generated from the same content the lessons use, so the workbook can never drift from the course
- Narrated video for every lesson — branded slides and a UK voice, rendered locally and automatically, with no per-minute video bill
- On-demand seminars built from real speaker notes, so the narration is the actual delivery rather than a script written after the fact
- An insight feed of business models, each with an original diagram drawn for it
- An admin panel for learners, enrolments, revenue and completions
- A full test mode so the entire platform runs and can be demonstrated before any payment or database keys exist
Extract properly, don't summarise
The first pass produced tidy summaries of each course. They were useless — a six-lesson skim of a full-day course is a brochure, not training.
The extractor was rebuilt to read each manual's own table of contents as the lesson map and pull the full body text underneath it. One course went from six lessons to thirty-seven, and from a summary to the actual material somebody paid for.
Workbooks generated from the lessons themselves
Every course has a branded learner manual, built from the same content the site serves.
Which means the workbook can't drift from the course. Hand-maintained companion documents always diverge eventually; generating both from one source makes divergence impossible rather than merely unlikely.
Video rendered locally, not bought by the minute
Branded slides plus a UK voice, rendered on a machine we control, automated end to end.
At roughly 1,600 lessons, per-minute video pricing would have been prohibitive — the library would simply not have had video. Owning the pipeline changed it from a budget question into a compute question.
A full test mode from day one
The entire platform runs on a local mock — signup, purchase, enrolment, progress, certificates — with no payment or database keys.
So it could be demonstrated and reviewed before a single account existed, and it switches to the real backend when keys are added. Nothing was blocked waiting on someone to open a Stripe account.