Liulishuo

Somewhere right now, a student in a third-tier Chinese city is talking to her phone. She reads a sentence aloud. A green meter twitches. A word lights up red - the vowel was flat. She says it again. The meter climbs. There is no teacher in the room, no classmate to wince at the mistake, no clock running down a billed hour. There is only the app, listening. That app is Liulishuo, and it has been having this exact conversation, in some form, with more than a hundred million people.

The name translates to "speak fluently." It is the kind of name that sets a trap for its owners - promise fluency and you are judged on fluency. Liulishuo built the whole company around that single, stubborn verb: speak. Not memorize. Not pass. Speak.

A nation that could read English but couldn't speak it

China had a peculiar English problem. Hundreds of millions of people had studied the language for years - grammar drills, vocabulary lists, exams that rewarded silent precision. And yet most of them froze the moment they had to open their mouths. The classrooms were full. The speaking practice was empty. Good spoken-English teachers were scarce, expensive, and concentrated in big cities, which is a polite way of saying that fluency was, for most people, a question of postcode and budget.

You could solve that the slow way - train more teachers, build more schools, hope geography cooperates. Or you could decide, as three engineers did, that the bottleneck wasn't teachers at all. It was that no human teacher can sit with 100 million students at once and patiently correct every wobbly vowel. Software can. Software, it turns out, is an infinitely patient listener - which is more than most of us can say for our language tutors.

Three engineers, one apartment, a microphone

In 2012, in an apartment in Hangzhou, the company began. The founders did not look like typical education entrepreneurs. They looked like a speech-recognition research group that had wandered into the wrong industry - which was rather the point.

Their bet was that pronunciation - the part of language learning teachers find tedious and students find terrifying - was actually a tractable machine problem. A computer could compare your spoken sounds against a model of "good English" and tell you, instantly and without judgment, exactly where you missed. To do that well, they needed data: not a little, but oceans of it. So they collected recordings. Billions of utterances, eventually, each one a learner reading a line aloud, each one teaching the model a little more about how non-native speakers actually sound.

An AI that listens back

The flagship app, English Liulishuo, launched in 2013. Within months it climbed to the top of the Chinese App Store, which is the sort of thing that happens when you give a frustrated, exam-weary population the one thing the exams never offered: a safe place to be bad out loud. By 2014 the app was being featured on iPhones in Apple's China stores. By 2016 the company shipped the piece that gave it a name - the "AI English Teacher," a deep-learning engine that scored not just words but fluency, intonation, and comprehension.

What you can actually do with it

Open the app and you get a course that adapts to you. Read sentences aloud and it scores your pronunciation in real time. Get a learning path that reshuffles based on where you stumble. The product family fanned out from there: Liuli Reading turned articles from publishers into guided lessons with quizzes; IELTS Liulishuo drilled speaking tests for the exam-bound; courses for kids and standardized tracks followed. The throughline never changed - point a microphone at a learner and give back something useful in the time it takes to breathe.

The numbers behind the noise

Plenty of edtech companies promise scale. Liulishuo had the receipts. By the middle of 2018 it counted more than 80 million cumulative registered users, and learners eventually spanned 175-plus countries. Investors noticed: a Series B in 2015 from a who's-who of cross-border venture firms, a reported nine-figure Series C in 2017 led by China Media Capital, and then, on September 27, 2018, the loudest receipt of all - a New York Stock Exchange listing under the ticker LAIX that raised about $72 million at a market cap near $600 million.

The company also reached beyond its own apps. It set up an AI Lab in Silicon Valley in 2017, pulling speech and language research closer to the source. It packaged its engine for businesses through LingoChamp Enterprise, and partnered with hardware makers like Timekettle to push its assessment tech into translation devices. The pitch to partners was the same as the pitch to students: we already know what good English sounds like - plug in.

Popularize English - the verb does the work

Strip away the funding rounds and the ticker symbol and the mission is almost embarrassingly plain: make learning English cheap, personal, and available to anyone with a phone. The radical part isn't the goal. Everyone in edtech says some version of it. The radical part is the method - the insistence that the scarce, expensive ingredient in language learning, a teacher's full attention, could be manufactured in software and handed out for the price of a subscription.

It has not been a frictionless ride. The company's stock took the public-market beating that many Chinese ADRs did, and by the early 2020s it was fielding going-private proposals - a reminder that being right about the technology and being rewarded by the market are two different exams. But the underlying conviction never wavered: the bottleneck was attention, and attention could scale.

The patient machine was early

Liulishuo made its bet years before "AI tutor" became a phrase every startup deck used twice. It collected the audio, trained the models, and shipped a product that did one unglamorous thing extremely well - it listened to ordinary people speak imperfect English and helped, without flinching. That is the version of AI in education that holds up: not a chatbot that does your homework, but a tireless coach that catches the vowel you keep flattening.

So return to that student in the third-tier city. Before Liulishuo, her options were a crowded classroom that never let her speak, or a private tutor she couldn't afford. Now she has a teacher in her pocket who will let her say the word wrong fifty times and never once sigh. The meter twitches. The word turns from red to green. She moves on. Multiply her by a hundred million, and you start to understand what the company actually built: not an app, but a way for a country to finally open its mouth.