Jarvis — Turning Classroom Data into Teacher Decisions

Context

In 2021, teachers in the US were doing two jobs at once. They were teaching, often through a screen, and they were configuring software that had clearly been built for somebody else: administrators, IT departments, instructional designers, anyone but them. RIIID Labs had spent the year talking to enough teachers to know there was a product opportunity sitting in plain sight. Existing LMS dashboards treated teachers as data entry clerks. The thesis was simple: build the dashboard a teacher actually wants to open in the morning.

The problem

Teachers needed four things from a dashboard, and the existing tools delivered none of them well.

First, a fast answer to "how is my class doing?" Most existing tools answered this with a wall of metrics. None of those metrics added up to a decision.

Second, a way to spot the students who needed attention before they slipped further behind. The information existed in every LMS, but it was buried inside per-student profiles, several clicks deep.

Third, a coherent picture of where assignments stood. Teachers were juggling work across grading, submitted, and not-yet-due states with nothing more sophisticated than mental tally marks or sticky notes.

Fourth, an onboarding flow that didn't require an IT ticket. Most teachers wanted to import their roster, set up classrooms, and start teaching the same day. Some of them did not have an IT department to call.

My role

I led the product design work from the start. The first three months I was largely solo on the product surface: framing problems, running research, sketching flows, building the interaction layer. The illustration, motion, and visual identity work ran in parallel and scaled up as we moved towards build, bringing the team to three designers across product and visual disciplines. My remit stayed on the structural side: information architecture, flows, components, and the decision logic that connected the data layer to what the teacher actually saw on screen.

Research

We ran eight teacher interviews and two administrator interviews across elementary school settings in the US. Personas weren't the deliverable. The deliverable was a journey map that connected what teachers said they were trying to do (teach) with what they were actually doing (logistics), and where those two things were in conflict.

A pattern surfaced quickly. Teachers didn't trust dashboards because dashboards routinely surfaced information that didn't lead anywhere. Show me a student is struggling, and then what? Show me the class average, and then what? The real design problem was making information lead somewhere from inside the surface where it appeared.

Dashboard

The home dashboard answered two questions teachers asked first thing in the morning: how is the class doing, and who needs me today.

The first half of the surface gave class-level performance at a glance. Not a chart for charting's sake. A single line that said the class is on track, or the class is behind on these two topics, with one click to go deeper.

The second half was the Needs Attention widget, the part of the dashboard I'm proudest of. We surfaced four to six students who were trending in the wrong direction. Not by aggregate score, but by recent behaviour: late submissions, missed assignments, declining accuracy on practice. Clicking a card opened a side panel with the student's recent work and a few quick actions: send a message, flag for grading, mark resolved. The widget was designed to make the next decision easier, not just visible.

Assignment status

Teachers were managing assignments across three states: needs grading, in progress, not yet due. We built a board layout with three columns and one rule: every assignment had a clear next step.

Moving an assignment from "needs grading" to "graded" wasn't a status update buried in a dropdown. It was a drag, or a click on the card itself. Teachers told us they wanted assignment management to feel like clearing a desk, not filing paperwork.

The assignment creation flow ran in parallel. We collapsed the standard six-step new-assignment wizard into a single screen with progressive disclosure: the core fields visible, the rest accessible but not required.

Clever integration

Most US K-12 districts use Clever for roster management. Our integration design started from a question: what's the smallest amount of work a teacher should have to do to get a working classroom?

We pulled rosters automatically, surfaced any reconciliation conflicts in plain language, and gave teachers a single approval step. Setup that would have taken twenty minutes through manual data entry took less than two through Clever sync. We did the same for security and SSO, both invisible to the teacher but loud in the procurement conversation.

Validation

We ran an unmoderated Maze test with ten participants on the mid-fidelity prototype. The test covered two flows: signing up and setting up a classroom, then viewing class and individual student performance.

The setup flow held up well. The performance views surfaced a useful insight: teachers consistently went to individual student records before checking class averages. We had built the dashboard the other way around. The fix wasn't to flip the layout. It was to make the Needs Attention widget more prominent at the top of the page so it could carry per-student information forward without forcing teachers through the class summary first.

What I took from this project

Jarvis validated the product hypothesis: a teacher-first dashboard built around decisions rather than data could clear specific friction from the daily workflow. The product moved into engineering build-out on the foundations we shipped, and the research patterns held up in subsequent rounds.

The lesson I carried out of this project, and one that shapes how I work now, is that AI-assisted workflows live or die in the surface that wraps the model. The model can predict performance, score practice, and flag anomalies. None of that helps a teacher who can't find it, doesn't trust it, or can't act on what it says.