A lecture PDF goes in; a proposed pathway of lessons and practice comes out. Here is how AdaptIQ’s AI authoring works, step by step — and why the teacher has the final say.
Most courses already exist in some form: lecture notes, slide decks, handouts. What takes time is turning them into something students can practise with — lessons cut to the right size, practice sets, exercises with explanations. AdaptIQ’s AI authoring is built to do the drafting, while the teacher makes every decision.
Step 1: the material
A teacher uploads source material: PDF, Word, PowerPoint, Excel, CSV, Markdown or plain text, JSON, or images. AdaptIQ extracts the text, indexes it and makes it searchable, so the assistant can find the right pages later. Images are described by a vision model, so a diagram is not a blind spot.
Uploaded files sit in a private bucket and can be downloaded only through an authenticated teacher endpoint. Whatever a document says, the assistant treats it as data: the uploaded text is fenced off in the prompt, separate from the instructions.
Step 2: a proposed pathway
When a teacher starts a new pathway with AI, the assistant reads the document’s outline and pages and proposes a structure: the pathway, its topics, and the lessons and practice sets inside each topic. Every step carries the page range it comes from, so the teacher can check the source at a glance.
Nothing is created yet. The proposal is a draft in the chat. The teacher can accept it, discard it, or ask for a change in plain words before anything becomes content.
Step 3: generating the content
Once the structure is accepted, “Generate content” writes it one step at a time: every lesson at 300 to 900 words, in the language of the material, and three to eight exercises per practice set. A progress bar shows which lesson or practice set is being written and what generating has cost so far. The teacher can stop it at any time; a step already running still finishes.
The exercises are not all multiple choice. The assistant can write ten of AdaptIQ’s twelve built-in exercise types — among them cloze texts, short numeric answers, ordering, sorting into categories, tables to fill in and concept maps — plus the generated statics problems of the biomechanics module.
Step 4: the teacher decides
Every draft is a proposal. A teacher can accept one, accept it together with everything below it, or accept a whole batch at once. They can discard what does not fit, or ask for a change — a change to an existing exercise is shown as before and after.
Before accepting, any exercise can be tried in the real student player, in a phone frame or a desktop frame, or full screen. Typing “@” in the chat points the assistant at a specific lesson, set or exercise, and “Ask AI to change this” on an exercise attaches it to the next message.
The assistant can also ask back: a clarifying question to the teacher, with quick-reply chips for the answer.
Step 5: review before students see it
Accepted AI content is marked as AI-made and set to “needs review”. Students cannot see it until a teacher approves it, and publishing is refused while unapproved AI content remains. Every save — by a teacher or by the AI — is stored as a revision, so an earlier version can be restored.
Costs and controls
AI authoring is off unless it is switched on for a site or a module. Each site has a daily AI budget in US dollars that fails closed: if the setting is malformed, AI work stops rather than running unlimited. Per-teacher rate limits apply, and the models — third-party large language models reached through OpenRouter — can be changed per site.
The result is not a course written by a machine. It is a first draft written fast, checked by the person who knows the subject, and published only when they are satisfied.