From a blank LMS to a launch-ready school in minutes
Designing an AI-guided onboarding and visual identity system that turns a teacher’s intent into a reviewable starting setup instead of an empty SaaS account.
Some product visuals have been simplified or reconstructed to protect confidential platform and customer information. School data is fictional.
An empty dashboard was asking teachers to become LMS experts.
New admins landed on the existing Get Started experience and still had to create courses, quizzes, websites and the rest of their setup by hand.
The problem wasn’t education. It was time-to-value.
The system needed to help a new educator reach a meaningful starting state quickly enough to understand Learnyst’s value, while still giving them control over what eventually goes live.
“The first experience needed to demonstrate what the platform could become — not teach every feature inside it.”
“I don’t know what to create.”
“Users need to reach value earlier.”
“We have many products and cannot build one onboarding for every teaching segment.”
Before designing onboarding, I needed to understand what “getting started” actually meant.
Don’t explain the platform. Give the user something worth editing.
Don’t explain the platform. Give the user something worth editing.
Instead of a checklist telling admins to create a course, build a website and configure branding, the new experience generates a usable draft setup from a small amount of context.
Three questions were enough to establish useful context.
Enough context to personalize. Not enough questions to turn onboarding into another setup form.
AI creates momentum. It doesn’t take ownership.
Generation produces a starting point. Every decision that makes something public stays with the admin.
Generating an initial structure from the teaching context it was given, proposing a visual identity, and producing course and sales-page starting content.
Review, edit, selection, customization, and the publish decision. Generated is never published.
Generate first. Explain through the result.
The flow is linear. Every step except content generation can be skipped, and a skipped step stays available afterwards.
One step is mandatory. The rest wait for you.
Generating content wasn’t enough. The school also had to feel intentional.
Teachers were creating logos, themes, thumbnails and assets by hand, which made presentation inconsistent across the school. One school now resolves to one theme object.
Personalization needed boundaries.
The AI is not choosing from an unlimited visual space. Designers define the safe visual possibilities; the system selects and composes within them.
Generation needs a deterministic fallback, not only a good day.
Every school needs a mark. The logic never leaves that to chance.
One identity had to survive more than one screen.
The platform already had theming concepts across several surfaces. The design problem was making a generated identity behave as a system rather than as unrelated image generation.
Designing for a real platform, not a clean prototype.
A setup that assembles several products at once will sometimes half-succeed. The states below are the ones I designed around, and they decide what a retry is allowed to touch.
I translated the UX states into implementation behaviour, kept existing entities and frameworks in play, and validated the result through design QA.
The platform was never a blank technical system.
Success meant reaching value faster — not completing a tour.
The numbers below are the targets and the measurement plan the work was designed against. None of them is presented as an achieved production result.
A product target the work was designed against. No measured production figure is claimed.
A target stated in one documented plan.
A target stated in that same plan.
- Onboarding step completion
- First content accessed
- Profile completion %
- App vs web
- Time to first lesson
- Generated-product engagement
- Support and help demand
From guidance to a generated starting point.
Instead of entering an empty setup experience, a new admin can move through AI-assisted onboarding, generate a starting product setup, review how the school looks across surfaces, and continue into customization and publishing.
What this changed in how I design AI products.
AI works best when it eliminates the blank state.
Generation needs deterministic boundaries and fallback behaviour.
The strongest AI experience still needs a clear review, edit and publish model.
And the generated output teaches the product better than a tour: it gives the user something real to explore.
What I’d measure and improve next
- Compare activation across teaching segments
- Measure edit depth on generated output
- Track signup to first published product
- Accessibility validation of automatically selected palettes
This project taught me that an AI feature earns trust by producing something worth editing, not by explaining itself.