Jen Liu and the grandmother who became an English tutor

The grandmother came first. Before Toko became an English practice app, Jen Liu and Erica Du had a smaller, rather charming project: an artificial grandmother with whom they could practice Mandarin. They called it GPT 奶奶. Plenty of software begins with a business plan. This began with someone to talk to.

Liu later described the experiment as something the pair had “hacked together.” They used GPT-3, then followed the demand toward English learners in East Asia. An idea built for their own language practice found its commercial direction in people practicing a different language. The original grandmother had acquired a surprisingly international classroom.

There is a temptation to let that anecdote carry the entire story. It is a good anecdote. But Liu had been working on educational software long before the AI grandma appeared. Look backward through her career and you find questions written into textbooks, vocabulary words falling through space, and the decidedly unglamorous work of making a screen respond properly. Toko belongs to that longer history.

Before the chatbot, a question in the margin

At MIT, Liu studied computer science and engineering. During the 2013-2014 academic year, she was an MIT EECS | Google Undergraduate Research and Innovation Scholar, supervised by David R. Karger. Her project had an ambitious title: The Future Textbook: Collaborative Annotations and Inline Exercises.

The work concerned NB, a tool for asking and answering questions directly on PDF textbooks. A student could encounter a difficult passage and discuss it where the difficulty occurred. Teachers, in turn, could see what students understood and adjust their teaching. The margin had a job to do.

Liu proposed inline exercises, shared annotations that could consolidate a discussion, and ways for students to rate the helpfulness of contributions. These were specific mechanisms for making reading more participatory. A textbook could become somewhere to try an answer, see another explanation, and leave evidence of confusion before the class moved on.

She had also worked on mobile educational games at MIT's Scheller Teacher Education Program Lab. Even then, her interests included the practical experience of learning through a device. The subject matter would change. The attention to what a learner could actually do on the screen would remain relevant.

A short journey from Newton to outer space

Liu grew up in Newton, Massachusetts, and attended Newton North High School. In August 2014, Quizlet welcomed her as a full-time engineer after two internships. She had graduated from MIT that May. The company's welcome offered a pleasantly ordinary beginning: a returning intern joining a team whose product she already knew.

By February 2015, she was back at MIT with colleagues Jeff and Arun for the university's xFair, recruiting engineers for Quizlet. It is a small moment in a career, but a useful one. The former student was now helping her employer find its next builders. Education software involved people on both sides of the recruiting table.

Then came the asteroids. Quizlet's Gravity game sent vocabulary words falling toward a planet. Players had to type the right answers before impact. Forget a word and the consequences became astronomical, although mercifully confined to the browser.

In a 2016 engineering essay, Liu explained how the team made those animations work in React. Falling asteroids, advancing planets, and a level-up badge each needed different treatment. The engineering was in service of something immediately legible: a learner should understand what was happening quickly enough to keep playing.

The work behind a falling word

Liu's accompanying animation presentation is revealing in its specificity. It compares approaches, considers their tradeoffs, and finishes with a reminder that building takes iteration and performance testing. A feature that looks effortless to a student can require a roomful of engineering decisions.

One approach handled the downward motion of asteroids. Another handled planet movement. CSS animation handled the badge. Liu did not present a single technique as a universal answer. Different behaviors called for different tools. The visible result was a game; the underlying work was a series of choices about timing, control, and performance.

This matters when considering her later work with AI. Educational products still have to turn technical capability into an experience someone can use. The learner encounters the sentence, the correction, the next prompt. Whether those moments are clear and well timed remains a product question, however sophisticated the machinery behind them becomes.

Liu went on to become a senior product manager at Quizlet, with work spanning AI-powered learning, mobile subscriptions, and international growth. Engineering gave her experience with how a feature behaved. Product management broadened the responsibility to which features a learning business should build, support, and bring to new audiences.

A concierge before a conversation partner

Between Quizlet and Toko, Liu founded Uplevel, a coach-matching concierge service. She ran that venture in 2019-2021. It is a different kind of business from a language app, and a useful interruption in any tidy story that makes one job look destined to lead to the next.

A concierge service and an AI tutor arrange help in different ways. One matches a person with a coach; the other makes practice available through software. Both put the problem of finding suitable help near the center of the experience. The resemblance offers one way to read Liu’s career: an interest in how people reach the help they need, even as the means of providing it changes.

By 2021, Liu and Du had founded Toko. The company joined Y Combinator's Winter 2022 batch. Taiwan was its first market. Its initial description was concrete: short conversations with AI, realistic topics, and grammar feedback for English learners in East Asia.

Toko's stated ambition was to make language fluency accessible beyond people who could afford individual tutoring. That gives the technology a practical assignment. A conversation partner has to be available often enough, and at a manageable cost, for practice to fit into an ordinary life.

Making room for another attempt

The product's design puts speaking practice into familiar situations, from small talk to workplace discussions. A learner gets a chance to respond and receive feedback. The attraction is easy to understand: you can rehearse a conversation before you need to have it with somebody whose answer carries consequences.

Toko now offers more than 1,000 topics and controls for speaking speed, accent, and practice time. It also describes feedback on grammar, pronunciation, and expression. A learner can adjust the conditions of practice, then examine the response to an individual sentence. The details show where the team has chosen to spend its effort.

Work, travel, and exam preparation appear among its learning contexts. Together they give the app a fairly recognizable map of adult language needs: explain yourself at work, get through an unfamiliar situation, or prepare for an assessment. The individual sessions may be small. The occasions they prepare someone for can matter considerably.

Coffee, feedback, and the people using it

The company's public life also includes an invitation from Liu and Erica to meet for coffee in Phoenix. The Coffee & Conversation meetup offered visitors a chance to meet the founders and engineering team and try the app. Coffee and three months of Toko Premium were the thank-you.

There is something pleasingly literal about a conversation app arranging a conversation. A product can be used across countries while its makers still benefit from sitting across a table from someone. An invitation to test software over coffee is a modest event, but it makes the relationship between builder and user unusually visible.

The invitation addressed learners in Taiwan and Hong Kong. Liu is based in Brooklyn, New York. That geographical distance is part of the operating context: a team based in New York building a product for people whose learning routines, languages, and daily situations differ from its own.

The grandmother gets a return visit

In November 2025, Liu announced Toko for Mandarin. The founders had been using an internal version for roughly a year, she explained. It helped them prepare for user interviews and communicate with family. The people building language practice software were using it to practice language themselves.

She credited Kenny Wong with driving the project and invited people to try it and give feedback. The launch also returned to the original GPT 奶奶 experiment. English had supplied the commercial direction; Mandarin remained part of the founders' own reasons to build.

“It's helped us prep for user interviews, build user empathy, and communicate better with family.”

Jen Liu, announcing Toko for Mandarin

That list travels from the professional to the personal in a single sentence. Interviewing users and talking with relatives can both depend on finding words that arrive a little too slowly. For Liu, the app's usefulness included both occasions.

Her career now offers several versions of an invitation: write a question beside the paragraph, type an answer before the asteroid lands, speak a sentence and see how to improve it. The interfaces have changed from annotated pages to games to AI conversations. Each makes participation possible in a particular way.

The grandmother gives this story its memorable beginning. The return to Mandarin gives it a satisfying turn. After years of building ways for other people to learn, Liu was still willing to occupy the learner's seat. There was another conversation to prepare for.