Turn any document into training, and prove people understood it

The training agent is not a compliance tool with other uses bolted on. It drafts from whatever is in your company brain, so the same process covers a new engineer learning your tech stack, a sales and support team getting ready for a product launch, a new standard operating procedure on the shop floor, customer service standards for a contact center, or anti-money laundering rules for a bank's front line.

You give the agent a title and a learning objective. It reads the relevant documents, proposes an outline of modules, and writes each module as a sequence of blocks: text, slides, audio narration, a multiple-choice quiz, a short scored scenario, or a role-play conversation with an AI-played customer. Admins can paste in video embeds of their own. The Understanding Check and its scoring rubric are written in the same run, so the check tests what the module actually taught.

• Every generated block records the passages it was drawn from, and a reviewer can open those sources beside the draft.
• A separate critique step scores each block against the module's objective and the source material. A claim the source does not support counts against the block, however well written it is.
• A block that fails critique is retried with the reasons fed back in. If it fails twice, it is marked for human review rather than quietly dropped or quietly passed.
• If you want to approve the structure first, the agent can stop at the outline and wait.

The critique step is a filter, not a sign-off. A program cannot be published until a person with admin or maintainer access has approved or edited every block in it, including blocks the critique step already passed. The app shows which blocks only the AI has checked and which a person has approved. Any block can be edited by hand or regenerated on its own, and the Understanding Check's prompt and rubric are editable.

An Understanding Check asks the learner to explain what they would do and why. The answer is graded one rubric criterion at a time, with feedback on each, so the learner knows exactly what was missing. Where an admin has mapped a module to a skill, a pass records a skill level on the person's record, the same skills your job roles require.

• A failed attempt is followed by a short re-explanation of only the missed points, drawn from the module's approved content, not the same page again.
• After three failed attempts the learner can move on, but the module is recorded as force-unlocked and gives no skill level, so a manager can see it was never demonstrated.
• A weak pass, one that took more than one attempt or left something unmet, schedules short refresher questions three days and ten days later.

The AI assistant can take training actions for you, calling the product with your own permissions. How much it does without asking is set per area on the autonomy dial, and the default is that it proposes and a person confirms.

Anti-money laundering and know-your-customer procedures, data protection, and health and safety are where training most needs to match your own policy and where a completion date proves least. Ascend drafts these programs from your policies like any other, and the record goes further than who clicked through.

• Assign by name with a due date, or by standing rule on job role, department and hire date, so new staff who match are assigned automatically.
• When training passes its due date, the person is notified and the escalation is written to the audit log. A weekly email rolls up each person's overdue training, training due that week, and refreshers waiting.
• For each person you can show every check attempt, the answer, the grading on each criterion, and whether a module was force-unlocked.
• Approvals, publishing, assignments and escalations are in an append-only audit log that admins can filter and export as CSV. When the AI assistant took the action, the log records the agent alongside the accountable person.

Does the AI decide who is compliant?

It grades written answers against a rubric your team can edit, and every grade explains itself criterion by criterion. What staff are taught is decided by the person who approves each block.

What happens when a policy changes?

Today, a maintainer regenerates or edits the affected blocks, and each goes back through human approval. Automatic flagging of training that cites a changed policy is not in the product yet.

Can we export completion certificates for a regulator?

Not yet. The evidence is the per-person attempt history and the audit log export described above.