IELTS Preparation Platform v2

01Why this, why now

IELTS preparation is bought by learners with a deadline and a number. They know the band they need and roughly when they sit the test, and almost every other decision — what to study today, which question type to drill, whether they are on track — is one they are unqualified to make alone. Most preparation products answer this with a content library and leave the sequencing to the student.

The bet in v2 is that sequencing is the product. Every screen in the prototype exists to answer one of three questions a student asks: where am I, what do I do next, and is it working. The content bank, the AI examiner and the gamification are inputs to those answers rather than features in their own right.

02Who it serves

The prototype is drawn for the deadline student and degrades sensibly for the others: an undated learner skips the date step in onboarding and gets a plan by band rather than by week.

03What success looks like

All targets below are proposed. None has a measured baseline in this product yet, since nothing is built; the first job after launch is to replace this table with real numbers rather than to defend these.

04Scope: brief to screen

Every line of the original brief, and where it is answered in the prototype. "Drawn" means the screen exists and is clickable with sample data; nothing in the prototype is wired to a real backend.

Out of scope for v2

• Native iOS and Android apps. Mobile web first; the layout is already built for it.
• Paid plans, checkout and entitlement. The prototype assumes every screen is reachable.
• Human marking of Writing and Speaking. AI only, with a teacher override path reserved for phase 2.
• Vietnamese interface. The prototype is English throughout; localisation is a later decision, not an afterthought to be bolted on.

05The band ladder, 3.5 to 9.0

The ladder is the spine of the product. It is twelve rungs — 3.5, 4.0, 4.5, 5.0, 5.5, 6.0, 6.5, 7.0, 7.5, 8.0, 8.5, 9.0 — and it does three jobs: it tags content, it gates content, and it reports progress.

Rules

• Everything carries a band tag. Every lesson, question set, mock and vocabulary list belongs to exactly one rung. Untagged content cannot ship.
• A rung unlocks at 80%. The rung below must be 80% complete before the next opens, so a student cannot skip from 5.0 to 7.0 drills and conclude the product does not work.
• Band is estimated per skill, not just overall. Overall is the rounded mean of the four, following the official rounding convention; each skill carries its own rung and its own path.
• Estimates decay. An estimate older than 30 days with no new evidence is shown as stale rather than as fact.
• Mock results outrank practice. A full timed mock moves the estimate further than the same number of untimed practice questions.

What a rung contains

06Personalisation and the daily plan

Onboarding collects five inputs, and each one changes something visible. If an answer changes nothing, the question should be cut.

How the daily plan is assembled

The plan is rebuilt each morning. It holds two to four tasks sized to the student's declared minutes, chosen in this priority order:

1. Notebook items due today, capped at one block so review never eats the whole session.
2. The weakest skill against target, measured as target minus current estimate. The prototype student is 2.0 below in Writing, so Writing appears first.
3. The weakest question type inside that skill, by accuracy over the last twenty attempts.
4. A scheduled mock, if the test date makes one due this week.
5. Something the student is good at, once a week, deliberately — a plan made only of weaknesses is a plan people quit.

Every task states its reason in one line: "Because your last essay scored 5.0 on Coherence". A plan whose reasoning is hidden is indistinguishable from a random content feed, and students treat it as one.

07Study by skill and the question bank

Each skill hub shows the current rung, the mastery of every question type inside that skill, and the band path. Mastery is accuracy across the last twenty attempts of that type, which keeps it responsive without being noisy.

Question type taxonomy

Filters the bank must support

The filter panel is the difference between a library and a practice tool. All eight dimensions combine, and the active set is shown as removable chips so a student can see why a list is short.

08Mock tests and official recent tests

Three products sit under one roof, and they are not the same thing:

• Official recent tests — released monthly, reconstructed from the most recent sittings, full four-skill format with exam timing. This is the headline: students search for exactly this.
• Full mocks — Cambridge 17 to 19, complete papers, the established benchmark.
• Skill mocks — one skill, 14 to 60 minutes, for a student who has an hour rather than three.

A completed mock produces a result report with an overall band, four skill bands, a breakdown of where marks were lost by question type, an updated plan, and a shareable card. The report is the moment the product earns trust or loses it, so it shows the student's own wrong answers next to the correct ones, with the passage or transcript that settles the argument.

09Speaking with the AI Teacher

Speaking is the skill students avoid, so the design goal is to make starting a session feel closer to a video call than to a recording exercise. The student picks an examiner — accent and manner differ — picks a part or a full test, and is then in a call: examiner on screen, their own camera in the corner, question asked aloud, live transcript running underneath.

Session flow

Scoring

Feedback reports Fluency and coherence, Lexical resource, Grammatical range and accuracy, and Pronunciation, each as a band with a one-line justification pointing at real evidence — "/θ/ produced as /s/ three times", "no complex sentences in the two-minute turn". A band without evidence is not shown.

10Mistake notebook and email reminders

One notebook, four sources, three kinds of entry. Everything the student gets wrong lands here without them doing anything.

Each entry keeps the word or rule, its IPA, its meaning, what the student actually wrote or said, and the sentence it came from. The wrong version matters: it is what makes the entry theirs rather than a dictionary line.

Review schedule

Spaced repetition with four stages — 10 minutes, 1 day, 3 days, 7 days — then mastered. A failed review drops the entry back one stage rather than to the start.

Email rules

• One email a day at the student's chosen hour, only when items are actually due. No due items, no email.
• It lists at most five entries, each with the sentence it came from, and carries one inline quiz question that can be answered from the email.
• An extra email is allowed when a mock adds five or more new mistakes, capped at one additional send per week.
• Every email carries a working unsubscribe and a link to change the time. Cadence settings live in the profile, not only in the email.

11Gamification, pride and ranking

Two different motivations are being served and they should not be confused. Gamification is private and keeps a daily habit alive. Pride is public and gives a reason to tell someone else.

Ranking scopes

Five boards, because one global leaderboard tells a Band 5.0 student only that they are losing: league, class, students with the same target band, friends, and global. The peer board is the useful one — it shows what students who started where you are actually practise.

12Training corner: IPA and pronunciation

A full chart of the 44 sounds — twelve monophthongs, eight diphthongs, twenty-four consonants — where the student's own weak sounds are marked from real speaking sessions rather than from a generic list for Vietnamese learners.

Each sound opens to: how the mouth makes it, a short mouth video, minimal pairs to hear the contrast, a record-and-compare with a score and a named failure ("/θ/ heard as /s/ in think"), and the words from the student's own notebook that contain it. Word stress and sentence rhythm are separate drills, because four-syllable stress errors and flat sentence rhythm cost Pronunciation marks independently of individual sounds.

The loop that makes this worth building: AI Teacher flags a sound, the sound appears in the training corner as weak, the drill uses the student's own vocabulary, and the next speaking session measures whether it moved.

13Chrome extension

Students who are ready for Band 7.0 vocabulary read English outside the product. The extension makes that reading count.

• Select or double-click a word on any page; a popup gives meaning, IPA, audio, the sentence it appeared in, its IELTS band level, and a synonym.
• Save it to a named list. It appears in the notebook and in the next review email.
• Optional: suggest words above the student's current band on a page, and highlight words already in their notebook when they read.
• A keyboard shortcut saves the selection without opening the popup.

14Teacher corner (Teado.ai)

Deliberately minimal, as the brief asked. The student side holds a class-code join and a placeholder explaining what appears once a teacher connects. Teachers themselves work in Teado.ai.

What phase 2 has to settle: assignments with due dates that auto-mark Listening and Reading and pre-mark Writing and Speaking for teacher review; a class ranking; a teacher override on any AI band, with the override visible to the student; and attendance. The hard part is not the UI. It is agreeing one band estimate that both the student app and the teacher app display, so that a student is never told two different things about their own level.

15Design system, mobile, accessibility

• Theme. Cloud theme, as set for this product: Inter, brand #6A309B, ink #1b1b1f, lines #e4e4e7, 8 and 10 pixel radii, flat surfaces, no drop shadows.
• One layout, two shapes. The app is a CSS container; under 720px the sidebar becomes a bottom tab bar, filter panels become bottom sheets, and two-column splits stack. There is no separate mobile codebase and there should not be one.
• Speaking needs the microphone. Permission is requested in context with an explanation, and a session degrades to text if it is refused.
• Audio needs captions. Listening transcripts unlock after submission; speaking transcripts are live.
• Colour is never the only signal. Status chips carry words, bands carry numbers, and weak sounds are marked with text as well as colour.

16Data the product must hold

Two fields carry more weight than their size suggests: the student's wrong version on a notebook entry, and evidence on a band estimate. Both are what make the product feel like it is paying attention rather than guessing.

17Delivery order and dependencies

Ordered so that each phase is usable on its own. A student could stop at the end of any phase and still have something worth opening.

18Decisions still open