Short-Staffed School, Full Course Catalog: AI Inside the Moodle You Already Run

There is a meeting that happens in a lot of schools around timetabling season, and it goes the same way everywhere. A subject comes off the list. Not because nobody wants it — the students asked for it, and somebody on staff would happily teach it. It comes off because there is no one left with hours to give it.

That meeting is the subject of this post. Not the version where a school buys its way out of the problem, but the version where the budget is what it is, the team is the team, and the question is whether anything can be done at all this term.

The shortage, in numbers

It helps to know the size of the thing you are standing inside. UNESCO's Global Report on Teachers puts the gap at 44 million primary and secondary teachers needed worldwide by 2030, with an estimated 15 million of those in sub-Saharan Africa. The report also records attrition among primary teachers almost doubling, from 4.62 per cent globally in 2015 to 9.06 in 2022, and prices the additional salaries at around US$120 billion a year.

The money side is just as stark. The World Bank and UNESCO's Education Finance Watch 2024 found that, as of 2022, annual expenditure per child in low-income countries amounted to no more than $55, and that total spending per child has largely plateaued even as total education spending has risen.

A teacher's week is not mostly teaching. It is drafting, sequencing, typing up question sets, building a rubric, marking, and answering the same six questions in six different messages. That is the part a tool can take a first pass at — and the part that decides whether a seventh course is possible or not.

The cheapest platform is the one you already run

The most expensive thing in edtech is rarely the licence. It is the migration — the term spent moving courses, the retraining, the two systems running in parallel while nobody trusts the new one yet, the enrolment data that has to be reconciled twice.

EduGears AI avoids that by not being a platform. It connects to the Moodle or Canvas you are already running over LTI 1.3, at the admin level, once. There is no second login for teachers, no parallel gradebook, and nothing to move.

• Setup is a site-level registration. About three minutes, with no plugin required. On Moodle, an optional plugin — local_edugears — makes registration one click when your site is reachable from the internet.
• A Moodle that is not reachable from the internet still works. If your server sits inside the school network, behind a firewall, or under bot protection, registration falls back to a claim code you enter by hand. This is a common setup, not an edge case, and it is documented rather than discovered.
• Content lands where teachers already look. Generated material is placed as activities inside your existing course sections, not in a separate portal students have to be told about.
• Grades go back to your gradebook. When an activity is graded, the score posts back into the Moodle or Canvas gradebook you already use. Practice sets, reflections, diagnostic quizzes and pages never post anything.
• You can take it back out. Modules export as SCORM 1.2, SCORM 2004 or Common Cartridge 1.3, and question sets as Moodle XML or QTI 2.1. Work you build here is not trapped here.

For a school with no spare IT capacity, that shape matters more than any feature comparison. There is no server to provision, no data to migrate, and no answer needed to the question "what happens to all our courses if this doesn't work out." You switch it off and your LMS is exactly as it was.

One teacher, one syllabus, one afternoon

The practical claim is narrower than "AI builds your course." It is this: a teacher who already knows the subject and has a syllabus can get to a reviewable draft of a whole course in a sitting, instead of over a term of evenings.

New Course takes the syllabus you already wrote — pasted in, or uploaded as a PDF, Word, PowerPoint, text or Markdown file. You choose how many modules you want, from one to twelve. The outline comes back in something like ten to thirty seconds, and you edit it before anything else is generated.

From there the working materials are generated per lesson, and each one is a draft in your workspace rather than something students can see:

• Lesson plans with timing and activities.
• Slide decks for the front of the room.
• Worksheets and study guides, which download as PDFs — a class set can be printed.
• Question sets: multiple choice, multiple select, short answer, open ended, essay, assertion and reason, statement based, and matching grids. Ten per run, and Generate More up to fifty.

If the material you want to teach from is not a syllabus but a stack of documents you have accumulated — a manual, a set of past handouts, an old scheme of work — Generate modules from files drafts outlines from those instead.

Rubrics: the AI drafts, you own the standard

This is worth being precise about, because it is the place where a shortcut would do real damage. The rubric tool proposes criteria and levels — between two and twelve criteria, two to eight levels — and shows them to you under a banner that says the draft has not been reviewed. Nothing is saved, nothing is attached to an activity, and no student is graded against anything until you have read every criterion and clicked Create rubric yourself.

Marking, drafted

Marking is where short-staffed schools lose the most recoverable time, and it is also where the most careless claims get made. Here is the actual behaviour, including the part that is not automatic.

On Student Assessment, an AI grade never reaches the gradebook on its own. It waits, labelled as ready for review, until a teacher opens it and posts it. If you change the score or the feedback, the result is flagged Teacher edited, so the record shows who decided what.

On project assignments, that is a choice rather than a rule: AI drafts, I approve holds the grade for you, and Fully automatic posts it straight to the gradebook. We would rather write that down than let a school discover it in week three. If your marking policy requires a human in the loop, leave projects on the approval setting.

The question at nine in the evening

A student stuck on question four at nine in the evening has, historically, two options: stay stuck until tomorrow, or ask a general-purpose chatbot that has never seen the course and will cheerfully hand over the answer.

Cog is the third option. It is a single Ask Cog button in the bottom-right corner of every page — or Cmd/Ctrl + K — and it already knows which lesson, quiz or resource the student has open, so nobody has to describe their own homework before they can ask about it.

• Learners get the tutor, and only the tutor. There is no persona toggle for a student — not a hidden one, no toggle at all, and the server enforces that regardless of what a browser sends.
• It withholds on unattempted work. On a practice question a student has not tried yet, Cog gives hints rather than the answer.
• It answers from your course. Replies draw on the materials the instructor uploaded to that course's knowledge base, so it is working from your sequence and your terminology rather than the open internet.
• Whether it appears at all is your administrator's call. Cog is configured at the institution level, so a school that does not want a tutor surface during a particular term is not stuck with one.

The honest framing is not "AI covers the evenings you cannot staff." It is that a student who would otherwise have lost an evening gets unstuck, and the teacher does not arrive on Tuesday to the same six messages.

The interface in your students' language

For a lot of schools the working language of the software is a real barrier, and it is one that quietly costs staff time in translation and hand-holding. The EduGears AI interface ships in sixteen languages:

English, Spanish, French, German, Italian, Portuguese (Brazil), Russian, Ukrainian, Arabic, Indonesian, Hindi, Telugu, Gujarati, Marathi, Simplified Chinese and Traditional Chinese. Arabic runs right-to-left throughout. An administrator can pin the whole institution to one language, overriding whatever the LMS sends, and optionally hide the in-app language picker so students do not wander into a locale they cannot read.

What AI does not fix

A post like this is worth very little without this section, so here is the list we would give a head teacher who asked us directly.

• It drafts. It does not know your subject better than your teacher does. A generated module in a subject nobody on staff can review is not a course you should be running. The tool shortens the drafting; it does not supply the expertise that checks the drafting.
• It does not supervise, notice, or hold a room. No part of this replaces a person watching a class and spotting the child who has gone quiet.
• It does not certify anyone. Whatever your regulator or ministry requires of a qualified teacher, it still requires. Staffing ratios are unchanged.
• It does not set your standard. Rubrics, pass marks, and what counts as good work remain your school's judgement, and every AI draft is presented as a draft for exactly that reason.
• It is a hosted service. Everything it generates, marks or answers happens against our service in the browser. The materials it produces — worksheets, study guides, pages — download as PDFs you can keep and print.
• It will not rescue a course nobody has time to teach. If there is no adult to run the lessons, a well-drafted module is a well-drafted module and nothing more.

What is left after all of that is still substantial: the first draft of everything, the marking pass, and the tutor a student can reach when their teacher has gone home. For a school choosing between offering a subject badly and not offering it, that is often the margin.

Where to start when budget is the constraint

You can evaluate the whole thing before any money is involved, which for a school working to a fixed grant is usually the only sequence that is possible at all.

• The free plan includes every one of the 26 AI tools, with AI included to evaluate them. No card, and no trial timer counting down.
• Learners are always free and unlimited, on every plan including the free one. Class size is never the thing that pushes a school onto a paid plan.
• Paid plans are sized by instructor seats — 5, 10, 15 or 20 — so what you pay tracks the number of people building and marking, not the number of children enrolled. The figures are on the pricing page.
• Or bring your own key. Connect your own API key for OpenAI, Anthropic, Google, DeepSeek or Sarvam, and AI calls go directly to your provider and are billed by them. Useful if your ministry or district already holds a provider account.

The order we would suggest: register the tool on a single course, hand it to one willing teacher with one syllabus, and look at what comes back before anyone writes a policy about it.

Frequently asked questions