Rather than redesigning the homepage from assumptions, I ran a series of generative research sessions with students: each one built their own "ideal homepage" from scratch in a shared whiteboard, with no existing design shown to them first. After a handful of these sessions, I synthesized the recurring patterns across participants into a single framework for prioritizing features, separating what students consistently asked for from what only one or two mentioned. That framework directly shaped what made it into the final Unified Navigation.

Teachers are stretched thin outside the classroom too, a lot of grading and progress-checking happens on their own time, squeezed between everything else in a day. That constraint shaped the dashboard and roster views: rather than digging through individual student records to figure out who's falling behind, a teacher can see cohort-wide status and clear next steps the moment they open the app.

Teachers also didn't share one workflow for organizing course materials. Some wanted to prepare an entire term's worth of readings upfront and sort them into weeks later; others preferred uploading materials incrementally, week by week, as each one approached. Rather than forcing a single pattern, I designed the upload experience, built together with Claude Code, to support both: teachers can drop files into a shared materials library and assign them to a week whenever they're ready, or upload directly into a specific week as they go. Materials can also be reassigned between weeks afterward without re-uploading, so an early organizational choice is never a dead end. Underneath that flexibility sits a fair amount of real complexity: per-file upload progress, deduplication so the same file isn't stored twice, and a background pipeline that feeds uploaded materials into the AI system used to generate assignments.

Grading was another consistent time sink — teachers don't have the bandwidth to write personalized feedback for every student on every assignment. That's where the system steps in: it generates draft feedback per student, and the teacher's role shifts from writing it from scratch to reviewing and adjusting what's already there.

Admin feedback made clear the different panels felt inconsistent to work in day to day, along with requests for both a grid and list view — some admins preferred a scannable table, others wanted a card-based view — and for the experience to work properly on mobile, not just desktop. That fed directly into a layout consistency pass across admin panels, the grid/list toggle on the class management screens, and a full responsive pass across the admin experience.

Since parents were a newer part of the platform, I ran individual interviews rather than assuming their needs from the student and teacher research. One clear theme: parents currently only get email digests of their child's assignments and performance, and those emails often felt like noise rather than signal, so they wanted updates reserved for genuinely important moments, not routine ones. They also wanted the same level of detail they're used to from a physical report card: specific subject areas where their child needs support, not just an overall score, along with a sense of where their child stands relative to their class and year. That shaped a clear job to be done: when a parent notices their child's performance dropping, they want to know exactly what action to take so they can step in early.