Education ecosystem
Three platforms for an education group with more than a dozen brands, on the web and as iOS and Android apps. I led their design from January 2025 to March 2026.
The client and product names are anonymized. Visuals are schematic recreations of the real interfaces.
TLDR
Context
The group sells courses, master’s degrees, university degrees and corporate training under more than a dozen brands. Three platforms sit behind it, sharing one design language.
The challenge
The three platforms are connected. From theirs, the tutoring team serves the students of the other two, and can see what each student sees, in read-only mode. Course content lives in Moodle, and the platforms wrap around it.
On top of that, all three share one design system, so every decision was multiplied by every brand, on the web and in the apps. And behind the scenes, the tutoring team was grading and routing tens of thousands of conversations by hand.
These are the four decisions that mattered most.
Decision 01
Inboxes piled up tens of thousands of conversations and you had to act on them one by one. Finding a student or a conversation meant filtering through a six-level hierarchy, the workload was uneven, and when someone left the team their conversations had no owner.
An interface built for volume: multi-select that lets you pick the page or all the content and processes it in the background; advanced filters ordered by the client’s real hierarchy; and automatic sharing-out based on each person’s ongoing conversations, with an “Available / Unavailable” status and offboarding that asks what to do with open work.
These were the foundational decisions of the tutor platform: what came later, such as AI-assisted grading, rests on an inbox that copes with the volume and a fair share of the work. In the filters, faculty and school only switch on once you pick an area, institutions combine with “AND/OR”, and quick counters don’t mix with filters, so they never give contradictory results.
Decision 02
With the inbox sorted, the next bottleneck was grading: the tutoring team reviewed every submission by hand.
The AI proposes a grade from the submission, the answer key and the course criteria. The tutor sees all three side by side and approves or edits it before sending it.
Grading isn’t fully automatic: the tutor still reviews every submission. In return, students never see a grade a tutor hasn’t approved, and there is always a manual path.
Decision 03
Every client wanted to see its own brand, colours and catalogue, and designing each brand separately couldn’t be maintained across more than a dozen.
One system with many identities: a shared base palette, primary and secondary colour scales per brand, variables with a mode for each, and translation notes in Spanish, English and Italian.
Clients upload their logo and choose from curated palettes, with no free-form colours. They lose some freedom, but the system stays one and every new brand comes from configuring it.
Decision 04
The student platform had grown by accumulation. The home screen, on web and app, was a shop window: a search bar and rows of catalogue with prices and ratings, and what the student was actually studying sat in a small strip. “My courses” didn’t cover everything they learn, and a subscription bought in the app couldn’t be managed on the web.
I redesigned the information architecture of the web and the app. The home screen now starts with the student: their courses in progress and their inbox, with the catalogue below as recommendations. Navigation groups everything else by what students need to do (learn, quick access, support and resources) and, in the app, “My learning” takes the centre of the bottom bar.
People who are already enrolled come in to make progress, not to buy. The catalogue is still there, but it no longer hides what someone is halfway through. To avoid breaking their habits, a temporary version let them move from “My courses” to “My learning” gradually.
Decisions made for mobile
Outcome
One design system serving 12+ brands across web, mobile apps and internal tools.
Work that used to be manual, such as grading, routing conversations and bulk updates, became assisted, automatic or done in a single step. Bulk actions are the clearest example: the team, who used to update conversations one by one, told us they now do it in a couple of clicks.
Told through the assessment screen, from requirements to the live version.
Every two weeks I met with stakeholders to gather requirements, and often with the tutors, who told us what they needed and where they got stuck. It all went into a shared glossary so design, engineering and the business used the same words.
I mapped the flows in FigJam, with open questions noted on the flow itself, and turned the ideas into low-fidelity wireframes, in Figma and often on paper.
The wireframes moved to high fidelity milestone by milestone, and each milestone was presented to stakeholders before moving on.
Every two weeks I presented the prototypes to stakeholders, along with how the sprint had gone and what had improved. I also tested with tutors, in sessions or through what they told me after using it, and that turned into usability improvements.
Before development I solved the edge cases: errors, empty states, unsaved changes or unpublishing with active enrolments. Around 70 modals and states by the third milestone.
I handed over linked prototypes showing the exact flows and every possible case, and wrote each task in Jira with its acceptance criteria, for engineering and QA. The team built it milestone by milestone, using the same glossary.
What I learned
AI-assisted grading works because the tutor still decides: the AI saves them the first draft.