Co-designing a 0-1 K-12 tool that helps teachers guide students to set and hit better goals

PLUS is a Carnegie Mellon tutoring program that pairs students with tutors during the school day to build math skills on i-Ready. Goal-setting was already part of it.

A student would sit with a tutor, agree on how many minutes and how many lessons they would do that week, and sign a paper contract. If they hit the goal, they got a fruit snack.

Goals were set verbally, written on paper, and then disappeared. Tutors ran the conversation inside a 17 to 21 minute session. Teachers got a spreadsheet once a week and were expected to do something with it. Students signed the contract and forgot the number by Wednesday.

The metric underneath all of it was minutes, and minutes could be faked.

Teachers know what fits inside a school day, what motivates a twelve year old, and which numbers actually change what they do on Monday morning.

Four longitudinal workshops with seven teachers, February to May 2026. One in person, three on Zoom. Each session built directly on the one before it.

One workshop ran in person at the school. Idea stations, printed storyboards, blank whiteboards. Before anyone saw a prototype, teachers sketched their ideal mid-semester report from nothing.

Rather than test static mockups, I built a functioning teacher dashboard in code and brought it into interviews live. Every teacher clicked through the real thing, and the prototype kept changing as interviews surfaced new friction.

01 · Built from findings, straight from the workshops Lessons passed as the primary metric, an attention banner at the top, a three week trend as the default view.

02 · Iterated across live sessions Each interview doubled as a design critique. Teachers pointed at specific elements and those became the next build.

03 · Closed the loop on onboarding Later interviews revealed teachers needed to configure goal parameters before the semester started, not during it. The onboarding flow was added and validated inside the same session.

Every goal started out purely weekly. Minutes and lessons, reset each Monday. It worked, but it left students with no sense of why any of those numbers mattered.

I made the call to connect the weekly micro-goal to a trimester diagnostic target, so the line between "pass two lessons this week" and "reach grade level by spring" was visible to both the teacher and the student.

Workshop transcripts, interview recordings, whiteboard photos and sticky notes were coded into a single framework. Every code carries a definition, a design implication, the participants who showed the pattern, the source file, and a paraphrased excerpt.

The point was to make each design decision traceable back to specific people saying specific things, rather than to a general impression of what teachers wanted.

Finding 1 · Minutes can be gamed Students idle or click slowly to run the clock. Tim: "Completing a lesson doesn't give me helpful data. Passing it does." Decision: lessons passed becomes the primary metric, and the gap between lessons completed and lessons passed surfaces everywhere as an integrity signal.

Finding 2 · Teachers cannot review 150 students Julie: "Rather than reviewing all 150 students, I want five rock stars to acknowledge and three to five kids who need immediate intervention." Decision: exception reporting instead of full rosters. An attention banner leads the dashboard, flagging students who are inactive, dropping, or stuck on the same lesson.

Finding 3 · Framing changes motivation In the student role-play, showing how far someone has come landed better than showing how far they are from grade level. Decision: the student view defaults to growth from starting point. Grade-level comparison is a toggle, not the default.

Finding 5 · Teachers want AI recommendations, not black boxes In the storyboard session they were positive about goal recommendations and immediately asked to see the reasoning. Decision: every recommendation shows its rationale, and nothing takes effect without teacher approval.

The research kept pointing at three moments where a teacher needs to be in the loop: the start of a semester, the week to week rhythm, and the individual student check-in. The tool maps to those three moments and nothing else. It is built and running.

Teacher onboarding Teachers set their goal metrics, reward triggers, and how much the system is allowed to recommend, before the semester starts.

Teacher dashboard Exception-first. The attention banner surfaces students who are inactive, dropping, or stuck on the same lesson. The roster groups by shared lesson so a teacher can pull the five students all stuck on one-step equations and run a five minute mini-lesson.

Student check-in report Shareable in one click from the roster. Current goal, three week trend, and skills led by strengths before gaps. Built-in conversation prompts so the check-in does not depend on the teacher improvising.

The vibe-coded prototype proved the concepts but it was not something a developer could ship. Once prototype testing settled the feature set, I built the design system the production tool was made from: 49 components and tokens.

Co-design looks loose from the outside. It is not. Every session ran on a written protocol with timed activities, specific prompts, and a defined thing I was trying to learn. These are the instruments the findings came out of.

01 · Workshop facilitation protocols Each workshop ran on a written protocol: session flow, timings, the research question behind every activity, the prompts as spoken, and a "watch for" list. What it drove: the journey-mapping prompt "what do you wish you knew but do not" is where the one day lag surfaced. Nobody would have reported it if I had only asked what they do.

02 · FigJam boards, blank and filled Both boards went in blank. What it drove: running them blank rather than pre-populated meant the columns filled with what teachers actually track, which turned out to be Classwise, walking the room, and a personal Excel sheet. That gap became the case for the tool.

03 · Thematic analysis on Dovetail and Taguette A nine column structure: code ID, name, definition, design implication, friction type, participants, source, excerpt, and open questions. What it drove: the design-implication column is the reason every decision traces back to specific people rather than a general sense of what teachers wanted.

04 · Storyboards Five illustrated concepts, annotated with four sticky colours for positive, question, concern, and idea. What it drove: colour-coding made concerns visible at a glance. The concerns on the AI storyboard are what produced the recommend-and-approve pattern.

05 · Feature prioritisation Must Have, Nice to Have, Not Now, sorted with teachers rather than for them. What it drove: grouping by shared lesson came out as Must Have while grouping by urgency and lesson time came out as Nice to Have. That ordering set the build sequence.