Learn words by living them — an AI-native vocabulary app built end-to-end, solo
Phraseq teaches words through contextual usage, not definitions. Where Anki shows you a card, Phraseq shows you 25 sentences tuned to your profession and register. Where Quizlet tests recall, Sentence Lab grades your writing and explains errors in plain English. Designed, built, billed, and deployed by one person.
| Industry / domain | Consumer EdTech · AI-native vocabulary |
| Platform | Web App — Next.js 14 · Supabase · Claude API · Stripe · Vercel |
| Timeline | 2026 |
| My role | Lead UX Designer & Founder — Strategy, Research, UX, Visual system, Full-stack build, Stripe billing, Deployment |
| Status | Live on Vercel — phraseq.vercel.app |
02
Flashcard tools are optimised for recognition, not production. The gap between knowing a definition and deploying a word naturally in writing is where learners stall — and where every major vocabulary app stops helping.
The standard vocabulary loop — see word, see definition, see one example, flip card — teaches recognition. You learn to match "ephemeral" to "short-lived." You don't learn that "ephemeral" belongs in literary prose and sounds affected in a work email. You don't learn the register.
For IELTS candidates, this gap is high-stakes. Band 7 and above requires demonstrating that you can use vocabulary flexibly in context — not just accurately, but appropriately. A Band 5 response uses words correctly. A Band 9 response uses them with precision and control. No flashcard app teaches the difference between those two things.
The core failure: existing tools treat vocabulary as a retrieval problem. It's a production problem. Users need practice reaching for words in context, not practice recognising them on a card.
03
Each decision is explained through the tradeoff it resolved, what was designed, and what it produced.
04
Every feature is a different attack on the same problem: get the learner using words, not just recognising them.
| Module | Core job | Key UX decision | Type |
|---|---|---|---|
| Word Vault | Enter any word, select your profession, get 25 AI-generated sentences across 5 registers tuned to your professional context | Profession selector is required before generation — register context is structural, not optional | Core |
| Sentence Lab | Submit a paragraph, receive a score, plain-English error explanations, and a suggested rewrite with changes highlighted | AI prompted to explain errors as a tutor would — not as a grammar checker. Jargon-free by design | Core |
| IELTS Bands | Submit a sentence, get simultaneous Band 5/7/9 rewrites with per-column vocabulary and structure annotations | All three bands shown together — learners need the contrast to understand what "Band 9" means in practice | Core |
| Quiz Engine | Auto-generated vocabulary quizzes derived from the user's Word Vault history — tests contextual usage, not definition matching | Quizzes generate from the user's own saved words — every quiz is directly relevant to their personal vocabulary | Engagement |
| Daily Words | One curated word per day with full Vault-style generation — seeded from the IELTS Academic Word List | Daily word is shown before the user reaches their saved vault — creates an entry habit separate from active study | Engagement |
| Streak Engine | Tracks consecutive days of active learning — vault lookups, Sentence Lab submissions, or quiz completions count | Streak counts any productive action, not just logins — prevents the Duolingo failure mode of opening the app to tap one button and leave | Engagement |
05
Every decision in this product is a design decision made visible through code. There was no engineering team to translate intent. What shipped is what I built.
Phraseq is the product I point to when the "designer who can build" question comes up — not because it demonstrates technical complexity, but because every UX decision has a direct implementation consequence I owned personally.
The BYOK model required encrypted API key storage in Supabase with row-level security policies. The IELTS band comparison required a carefully designed prompt returning three structured outputs in one API call. The Sentence Lab's plain-English feedback required a system prompt tested across hundreds of iterations to strip jargon without losing precision. These are design-build problems, not handoff problems.
Stripe billing is live. Users can subscribe, see their plan, and manage billing through a Stripe-hosted portal. The subscription unlocks the managed API tier. The BYOK tier remains free indefinitely as a retention-first acquisition model.
BYOK + managed subscription is a viable launch model for solo AI product builders who need to prove retention before absorbing AI infrastructure cost. The lesson generalises beyond Phraseq.
06
What Phraseq delivered — measured by product completeness, design quality, and the patterns it proved.
07
BYOK needs a no-auth demo moment first. New users hit an API key setup screen before they've experienced the product. The fix: let users run Word Vault once on a pre-seeded key before asking them to set up their own. Lower the first-value moment dramatically, then ask for the key. The current flow asks for commitment before delivering proof.
Streak needs a social share mechanic. The streak and daily word features are retention mechanics that stay entirely inside the app. The missing piece is a one-tap export of a Word Vault result or a Band 9 rewrite that a learner can post on LinkedIn or Twitter. Retention inside the app is valuable. Retention that becomes acquisition is how a solo product grows without a marketing budget.
"The vocabulary tools that exist are optimised for the test they were designed to pass — the flashcard recall test. Phraseq is optimised for the test that matters: can you write a sentence that lands? The gap between those two things is where I built. Designing and shipping this solo taught me that the most powerful design decisions are the ones you're also accountable for implementing."