How it works

Watch it build a course. Then watch someone prove they learned it.

Ten steps, in order, with the product running in every one. This is a demo manufacturer and its own product documents. Nothing on this page is a mockup.

The Creator agent lists the steps it took, including reading both ARC-7 source documents and the MEDDIC sales playbook, then produces a five lesson course with a time estimate on each lesson.
Part one, the Creator

Build the course.

One person who owns the training. No instructional designer, no authoring tool, and no starting from a blank page.

01

Add the documents you already have.

Product sheets, decks, process docs, spreadsheets. Drop them in or connect Google Drive or OneDrive. Everything gets read and indexed, and from this point on every lesson has to come from them.

Two ARC-7 PDFs being dragged into a new knowledge base and syncing.
02

It tells you where your own documents disagree.

Two files went in. Three contradictions came out. The technical data sheet prices the ARC-7S at $38,500 and the marketing overview says $36,500, and neither team knew.

You find that out here, on this screen, instead of finding out when a rep quotes the wrong number to a customer. Every line it writes carries the source it came from.

A knowledge base flagging three conflicts where two uploaded documents disagree on the ARC-7S price and the ARC-7H reach, with the synthesized text below carrying an inline source citation.
03

Add the way your team actually sells.

Upload your playbook and attach it to a job role. This one is MEDDIC. Pick from seven common methodologies or bring your own.

From here, everything built for that role is shaped by it. The practice calls, the grading rubric, the worked examples. Your method, not a generic one.

A MEDDIC sales playbook PDF being uploaded and attached to the Sales Representative role.
04

Ask for the course you want.

Type it the way you would explain it to a colleague. It asks what it still needs to know, proposes a course outline with a time estimate on every lesson, and waits for you to say go.

  • Reads your sources and your playbook before it plans
  • Shows every step it took to get there
  • Rewrites any lesson you push back on
A course being described in plain language in Creator Studio, and the agent replying with the problem, the audience and the outcome it has understood.
05

Publish to a role, not to a list of names.

Twenty five sales reps get it at once. Anyone who joins that role next month gets it automatically, without a person remembering to assign it.

Publishing the finished course and the confirmation that twenty five learners in the Sales Representative role now have access.
What it built

One set of sources. Five kinds of lesson.

The same two documents became a reading, an infographic, a podcast, a practice call and a written checkpoint. Nobody has to sit through a format that does not suit the work.

A generated infographic titled ARC-7P, plus or minus 0.03 mm repeatability, laying out the specifications as a diagram.
A generated podcast episode called Overcoming the Failed Automation Objection, with an audio player and its key takeaways listed underneath.
Part two, the learner

Prove they can do it.

One of those twenty five reps signs in and opens the course that was just published to her. She has a Mentor with her the whole way, and she does not get through it by clicking Next.

06

It asks two questions, then drops a lesson she does not need.

First sign in asks her job and what she is working on. That is the whole intake.

By the time the course opens, one of the five lessons is already struck out: "Covered in Sales Objection Handling". She proved those concepts in a different course last month, so this one starts at four lessons and thirty six minutes instead of five and forty one. Nobody sits through something they have already been marked on, and it is still there if she wants it.

What it asks
A short first-run form asking the learner for their role, seniority and current focus, saying it will be used to skip what they already know and pull examples from their field.
What it does with it
The course outline resolving from zero of five lessons done to one of five, with the infographic lesson struck through and labelled Covered in Sales Objection Handling.
07

Every lesson comes with a Mentor. Ask it anything.

She has an ROI case to build, so mid lesson she asks what labour rate to use and where that number comes from. It answers with $42 an hour, says the figure is fully loaded rather than base wage, and says which of the company's own documents it came out of.

She also has a drive tomorrow, so she asks to take the podcast next. It reorders the course and says it will come back to the reading when she is out of the car.

The lesson does not restart, and nothing she asks is held against her. Questions are never graded.

A learner asking mid-lesson which labour rate to use for an ROI case and where the number comes from, and the Mentor answering with a fully loaded rate of $42 an hour drawn from the company's own analysis.
08

Practice out loud, before it counts.

She sees who she is about to talk to, the three objections he is going to raise, and exactly what she is graded on, with the weights.

Then she talks. Out loud, in a real voice conversation, and it answers back in character.

The briefing before a roleplay: the situation, Joe the skeptical plant lead she will speak to, the three objections he will raise, and the graded criteria weighted 30 percent technical accuracy, 40 percent objection handling and 30 percent safety and compliance.
09

Miss it. See exactly what to fix. Pass it.

First attempt scores 48, and it does not fail her on a vibe. Technical accuracy comes back at 45 percent with the reason written out: she recommended the ARC-7S when the job needed the ARC-7P, and when Joe asked her directly for the repeatability spec she could not produce it.

Second attempt, same rubric, same grader. That same line comes back at 100 percent, because this time she cited the figure and the setup time the way the data sheet has them. Objection handling moves 55 to 82.

The improvement lands in the exact place the first attempt was marked down, which is the difference between a score and something worth doing twice.

Attempt one
A failed roleplay scoring 48, with technical accuracy at 45 percent and objection handling at 55 percent, and the reasoning naming the wrong model recommendation and the missing repeatability spec.
Attempt two
The second attempt against the same rubric, with technical accuracy at 100 percent and objection handling at 82 percent, the feedback noting she cited the repeatability figure and the four hour setup time correctly.
10

Finish on real answers, not multiple choice.

The course ends on a written checkpoint, two prompts and no options to pick from. She has to turn a plant's rework rate and staff turnover into metrics a CFO would act on, then name the model that fits and the next step in the deal. It is graded against a rubric built from the same documents the course came from: 90 for accuracy of ideas, 83 for completeness, 90 for clarity, and a note that the turnover figure never made it into her final business case.

Her answer
An open-ended checkpoint with two prompts, asking the learner to translate a plant's rework and turnover into MEDDIC metrics and to name the right ARC-7 model and next step, with their typed answers below each.
The grade
The graded checkpoint scoring 88, broken into accuracy of ideas at 90 percent, completeness at 83 and clarity at 90, with written feedback under each.

Then it closes the course out by name, tells her what she did well, and wishes her luck for the call tomorrow. Five lessons, one of them skipped because she had already proved it, thirty six minutes.

The Mentor closing out the course, marking the checkpoint at 88 percent and telling the learner she has completed Selling ARC-7 to Electronics Manufacturers and wishing her luck in the field tomorrow.
At the end of it

Something to show when someone asks if the team is ready.

A course that stays true

Built from your documents and tied back to them, so when a source changes you know what the change touches.

A score with the reasoning attached

Sub-scores per rubric line, next to the learner's actual words. Not a completion percentage.

The gaps, named

Who is confident and wrong, which lesson people keep failing, and who is not ready to take calls yet.

Questions this raises

Four that come up every time.

How long does the first course really take?

The run above is about fifteen minutes of work, and most of that is the wait while five lessons generate. The part a person actually does, adding sources, attaching the playbook, describing the course, is closer to five.

Does it invent things?

Lessons are written from the documents you upload and cite them line by line. When two of your own documents disagree, it stops and flags the collision rather than picking a winner, which is the third step above. Rubrics come from those same documents, so grading is judged against your material.

Do we have to build it ourselves?

Only if you want to. Send us what you already train on and we build the first course with you on a call, so you see it work on your own material instead of learning a tool first.

Can we see this on our own documents?

Yes, and it is the fastest way to judge it. Book a walkthrough and bring one document you already train people on.

Run it on your own documents.

Bring one document you already train people on and watch the same ten steps happen to it. Fifteen days free, up to 25 people, no card.

AI mentors grounded in your company's knowledge. Personalized upskilling for every role, from onboarding to mastery.
Based in Silicon Valley • Supporting remote teams globally
Navigation
Copyright © 2026 Mentor126.AI. All rights reserved.