Every AI-written lesson lands as a draft

Generating training content is the easy part. Deciding it is fit to put in front of two hundred colleagues is not, and that decision does not belong to a model.

Four ways in, one way out

A lesson can start from a topic, from your own PDF or Word document, from a real skill gap surfaced by a review, or from someone writing it themselves. Whichever route it takes, an AI-generated lesson lands as a draft.

A person reads it and publishes it. There is no configuration flag to skip that, because the failure it prevents is not a formatting error — it is teaching the whole company something that is confidently wrong.

What the model does not know

It does not know your reorganisation, your renamed teams, the policy you replaced in March, or the one client whose contract makes the general advice wrong. It writes plausible material, and plausible is exactly the failure mode that gets past a skim.

Documents are read, not kept

An uploaded document is split semantically into as many as eight lessons and is never stored. We take the structure and the content and let the file go — a source document is often the most sensitive thing in the exchange.

The quiz is not decoration

Every lesson ends with a multiple-choice quiz at a 70% pass mark, with unlimited retries and the best attempt counting. Unlimited retries are deliberate: the goal is that people end up knowing the material, not that we produce a clean distribution of first-attempt scores.