TeamLingua:為中文用戶打造的免費職場英語練習網站
I Built a Free Workplace-English Practice Site for Chinese Speakers
TeamLingua 是一款面向中文用戶的免費網頁應用程式,提供職場英語聽、説、讀、寫練習,毋須安裝;介面只有簡體中文。網站設有 500 題一般題庫,以及獨立的零基礎練習,包含 100 題、分為四個階段,避免初學題目影響一般練習的難度調整。用戶毋須建立帳戶即可試做一道題目。
I built TeamLingua in my spare time. It is a free web app for Chinese speakers who want to practise workplace English — listening, speaking, reading, and writing — in the browser, without installing anything.
One thing I want to say first: the interface is in Simplified Chinese. There is no language switch. I am saying that up front because this will not suit every reader here. I built it that way on purpose. The people I had in mind already read Chinese, and they need Chinese explanations next to English office language, not another English-only study app.
I am a Chinese student in Auckland. The gap I kept seeing was familiar: someone can pass an exam, then freeze when they have to book a meeting room or ask a colleague to review an email. A lot of existing tools are exam-oriented, single-player apps, or they ask you to download or pay before you can try a single question. I wanted a page you can open, try one question without signing up, and optionally come back with friends or colleagues to check in.
What I shipped
The whole front end is a single index.html hosted on GitHub Pages. Authentication and the database sit on Supabase. After you sign in, the app loads the main question bank from a Postgres table. Each row carries a skill (listening, speaking, reading, or writing), a difficulty level, and a JSON payload for the prompt, options, and the Chinese explanation.
The normal bank currently has 500 questions at levels 1, 3, and 5. The practice UI still presents a 1–5 level to the learner. Promotion and demotion use a rolling window of 15 scored answers: 80% or higher moves you up; below 40% moves you down. Same-day repeats and mistake-notebook reviews are excluded from that window, so grinding the same item does not inflate the level.
There is also a points system, a 20-minute daily goal, streaks, a mistake notebook, and a light team check-in so friends or colleagues can practise together. None of that is the interesting engineering part. The part I had to get right was the beginner track.
Why zero-basics is a separate track
I recently added a 🌱 零基础 ("absolute beginner") section: 100 questions in four stages — letters and common words, high-frequency words and phrases, simple sentences, and short office dialogues. You pass a stage by getting 20 questions right. After stage 4, the app offers to start ordinary practice at Lv.1, the lowest normal level.
The design constraint was this: beginner questions must not leak into ordinary practice, and they must not touch the level-adjustment window.
If those 100 items lived in the same bank, a beginner would be thrown into office-sentence questions too early, and a mid-level user who wandered into letter recognition would collect easy correct answers. Those answers would then sit in the 15-question window and could promote the ordinary level for the wrong reason.
So the beginner items are a separate track. They are marked as such, shipped in the page itself (the ordinary bank is fetched after sign-in; the beginner track still has to work without being mixed in), filtered out when the app assembles the normal bank, and skipped when a result is recorded into the level window. Points, streaks, and the mistake notebook still work the same way, because those are about habit, not difficulty.
The default tab is still ordinary practice. Only people who choose "almost zero basis" during onboarding land on the beginner tab first. That choice is remembered, but anyone can switch later.
Lessons
Keep adaptive difficulty and on-ramps apart. A rolling accuracy window is a simple way to adjust level in a single-page app with no recommendation service. It only stays honest if every item you feed it belongs to the same difficulty system.
Explanations should be in the learner's language. For this audience, a Chinese note on why "I want to order the meeting room" sounds off is more useful than an English UI.
Do not make the product depend on live AI generation. There is a button that can generate a fresh question, and it falls back to the bank if the request fails or times out. The 500-question bank, each item with a Chinese explanation, is what people actually practise with. I would not list this as an AI app.
A single HTML file is easy to deploy and easy to break. The beginner questions had to stay out of the ordinary path in more than one place: loading, rendering, scoring, and level adjustment. I would rather repeat a conservative filter than let one missed branch retune someone's level.
You can try one question on the site without creating an account. If you do open it, please remember the UI is Chinese. I would like feedback on the flow, the question design, and whether an English interface should come next.
來源:DEV Community · dev.to