Build the most convincing homework-help demo in the world: a clickable prototype where the homework visibly solves itself on the student's own photographed sheet, so anyone who picks up the phone says "whoa" within 10 seconds.
The others render answers in a chat. We perform intelligence: live, on paper, personal. Your job is to make that difference undeniable.
Not a slideshow. Not a spec. A prototype so real and so polished that we can hand a phone to a student, a parent, or an investor and let them feel the product before a single line of production code exists.
One sentence that defines the whole thing: competitors (Solvely, Gauth, Photomath) render answers in a chat. We perform intelligence, live, on paper, personalized. Everything you build serves that one difference.
The Job To Be Done is "make my homework be DONE." Everything else (learning, grades, parent trust) is an optional second layer that never blocks the student from finishing.
Today StudySmarter already ships a "Homework Helper," but as of the last audit (2026-08-25) it is a generic photo-to-answer chat wrapper with no link to our curriculum knowledge, student memory, or answer-verification engine. That is the exact commodity category every competitor is already in. This prototype shows what a genuinely differentiated homework product looks and feels like, before we build it for real.
If this demo lands, it becomes the reference artifact the product, design, and engineering teams build the real feature against. That is the impact of the internship.
The current prototype is good. "Best in class" clears a much higher bar. Use this as your definition of quality.
| Dimension | Today (v2) | Best-in-class target |
|---|---|---|
| First impression | Reads as a clean mockup | Reads as a real shipping app; a stranger can't tell it's a prototype for 30 seconds |
| The magic moment | Answers animate onto a stylized sheet | Answers write themselves in handwriting onto a real photographed worksheet, task by task, with live narration |
| Fidelity | Desktop phone-frame, demo data | Full-screen on a real phone, touch-first, feels native |
| Believability | One coherent worksheet | 2 to 3 believable worksheets and subjects, so a tester can't exhaust it |
| Personalization | One mistake-type chip | The app visibly knows this student: recurring mistake, next test, trajectory |
| Emotional payoff | A "done" screen | Relief and pride: 45 min to 6 min, a clean result, "you've got this" |
The blind test that defines "done": put this prototype next to Solvely or Gauth on two phones, hand both to 5 students, and ask which they'd actually use. Best-in-class means ours wins clearly, and they can say why in one sentence.
The prototype is a single self-contained file, index.html, in this same folder. No build step. Open it and it works. It has 11 screens across 3 layers, navigable from the right-hand rail or the arrow keys. You will grow this into the best-in-class demo.
The real product would need photo-to-task segmentation, an answer engine, structured output, and curriculum linkage. None of that is your job. You are faking all of it convincingly in the prototype. That is the point.
Extend index.html into a best-in-class demo. You own the prototype end to end. These are directions, not a checklist: the measure is the bar in section 03, not "did she implement my list."
If you find yourself wiring up a real service, stop. That is a signal you've left the scope.
Sized for roughly 6 to 8 weeks. Compress or stretch to fit the internship; the order matters more than the weeks.
Run the prototype, tap every screen, install and use all 3 competitors, write a one-page "what makes ours better and where v2 falls short" note.
Capture to "Live auf dem Blatt" is genuinely jaw-dropping on a real phone. This one screen alone should make people lean in.
Der Stift denkt mit, Fertig, and Heft-Seite polished to shipping quality. The full "get it done" arc feels like relief.
2 to 3 worksheets and subjects; personalization woven through so the app clearly knows the student.
Verstehen, Notenprognose, and the parent report elevated from mockup to believable.
Blind test vs a competitor with 5+ students; fix what confuses them; final polish pass.
Recorded walkthrough, a short deck of the wow moments, and the prototype documented so engineering can build the real thing against it.
Because this is a prototype and not a live product, the KPIs measure research depth, prototype fidelity, and validated desirability, not retention or revenue. The KPIs that matter most are in group C.
Understand the user. Complete 8 student discovery interviews (min 5, stretch 10) plus 3 competitor teardowns, by end of M0 to M3.
Land the magic. The hero moment triggers a strong positive reaction within 10 seconds for at least 70% of first-time users, by M1.
Beat the category. In a blind side-by-side, at least 80% of testers (4 of 5) pick our prototype over a leading competitor and can name why, by M5.
Feel like a real product. Reach a System Usability Scale score of 80+ (or 80%+ "I would use this weekly"), by M5.
Make it forwardable. Deliver a recorded walkthrough plus a wow-moment deck Chris can share without you present, by M6.
| KPI | Primary | Stretch |
|---|---|---|
| Student discovery interviews | 8 | 10 |
| Competitor teardowns, real worksheet each | 3 | 4 |
| Parent conversations (trust layer) | 2 | 4 |
| Teacher conversation (trust layer) | 1 | 2 |
| KPI | Primary | Stretch |
|---|---|---|
| Magic moments at shipping fidelity | 3 | 5 |
| Believable worksheets across 2+ subjects | 3 | 5 |
| Runs full-screen on real iOS + Android | Both | + tablet |
| KPI | Primary | Stretch |
|---|---|---|
| Moderated usability sessions | 5 | 8 |
| Unaided core-flow success (capture to done to result) | 80% | 90% |
| Blind preference vs a leading competitor | 80% (4/5) | 100% (5/5) |
| Time-to-first-"whoa" for first-timers | 10s @ 70% | 10s @ 90% |
| SUS score (or "would use weekly") | 80 | 85 |
| Spontaneous delight reactions | 1 per 2 | 1 per tester |
| KPI | Primary | Stretch |
|---|---|---|
| Recorded end-to-end walkthrough | 1 (max 3 min) | + social cut |
| Wow-moment deck | 1 (max 8 slides) | – |
| Weekly demo to Chris | Every week | – |
open homework-oneclick-prototype/index.htmlNo install, no server. Edit index.html, refresh the browser. To test on your phone, put the file on any static host or run a local server and open it on the phone's browser.
Ink write-on: animate max-width, not width:100%, or the checkmark gets pushed out of view.
The Unicode fraction slash (U+2044) renders invisible in some fonts; use a plain /.
For cross-fade between screens, give each screen a solid background.
index.html, tap through all 11 screens, arrow keys included.You do not need this to start building (section 10 is your start). It is here so you have the full picture and so this brief is the single source of truth. The strategy below is the real product and business context behind the demo; the last two parts are explicitly not your job.
The v1 to v2 lesson (read twice): the first prototype rendered intelligence as tidy UI cards. Chris's verdict was "doesn't feel a magnitude better than Solvely." The v2 fix: stop rendering answers, start performing intelligence. Answers ink onto the real sheet, the pen thinks out loud, the work happens in front of you. If a screen just shows a result in a card, it is not yet best in class.
The category today is per-task photo to answer. Gauth adds an "Emma" tutor and a library of ~100M solved problems. What none of them have, and what is our moat: curriculum knowledge, pupil memory, engine-verified answers, and teacher/parent relationships.
Per our notes, moonshots 1, 3, 4, and 5 are mostly wiring on assets we already have (see part D). For the demo, spend your "magic" budget on the ones that read as impossible for a competitor: 2, 1/3, 4, and 5.
Point-in-time findings from a prior session on the web-app repo, not re-read against current code today. Treat them as leads, not gospel, and confirm before relying on them. You do not touch this code; this is background.
systemMessageTemplate: 'homework_helper_conversation' (per the audit, ai-assistant-store.service.ts:401). Web: file upload instead.UNLIMITED_HOMEWORK_HELPER.'homework_tutor' is declared but never used: a free slot for the real tutoring flow.curriculum-lp.interceptor passes homework POSTs through with no bundled curriculum ids.CORE-637-homework-helper-improvements branch was dead; the helper shipped via CORE-663.This exists so that when your demo lands, the team already knows the path from demo to shipped feature.