Former ByteDance executives have started an overseas education project, using agents as "lifelong learning partners", and HSG has invested in it.
Text | Fu Chong
Editor | Su Jianxun
In November 2024, Li Kejia, Wu Jundong, and Zhang Qiming decided to start a business together to develop an AI Agent that "helps users learn throughout their lives." The new company was named Ouraca, which is an abbreviation of "Our Academy."
The three partners are all "veterans" in the field of Internet education:
Li Kejia, the founder and CEO, was formerly the founder of GeekData. After the project was acquired by ByteDance, he joined ByteDance's Smart Education business as the CEO.
Wu Jundong, a co - founder, holds a master's degree from Harvard Kennedy School and is an investor in TAL Education. He was also an investor in Li Kejia's GeekData project.
Zhang Qiming, a co - founder, was formerly the person in charge of ByteDance's education middle platform and has launched several products with tens of millions of daily active users and over 100 million monthly active users.
Ouraca was officially established in March 2025. Before its product was even launched, it had already received $7 million in seed - round financing. The list of shareholders includes VC institutions such as Sequoia China, Early Light Capital, and Etna Capital.
Interestingly, the above - mentioned investors all made very quick decisions when investing in Ouraca. Li Kejia recalled that on the night when he decided to start the business, he told Tian Jiangchuan, the founder of Early Light Capital, about his idea. Just a few minutes later, he received Tian Jiangchuan's reply saying, "Then I'll invest in you." Sequoia China also made its investment decision at a similar speed.
"Etna Capital made its investment decision almost within 24 hours," said investor Haina Xu, recalling the situation at that time. "Among the usage scenarios of ChatGPT, the education category has long ranked among the top five. It can be said that education is one of the industries that has been most changed by large - language models. People have been looking forward to an all - powerful AI Tutor for many years, and Ouraca happens to be the first team we've seen exploring the 'lifelong learning product in the AI era'."
In September 2025, Ouraca launched its first application, "Aibrary," which uses a unique "AI - generated podcast" feature called Idea Twin to guide and accompany users in learning. Currently, Aibrary and Ouraca's new products are still undergoing rapid and continuous iteration.
Using AI as a "Learning Buddy" That Asks Questions
Li Kejia has his own set of judgments about AI education products: Since large models can quickly generate a large amount of content, the pain point is not the lack of content, but how to help users ask good questions to AI and obtain high - quality answers.
On Aibrary, users can actively ask Aibrary questions about the topics they want to learn. If users don't know what to ask, Aibrary will push three life - related topics to users every day. For example, one recent topic was about "small habits that add up over time."
In each topic, Aibrary embeds its own insights, recommends a reading list worth reading, and also provides book explanations.
However, Aibrary is not just another "explain a book in X minutes" app. Instead, it turns knowledge into conversations that users are interested in and then presents them in the form of podcasts.
Specifically, the podcast has two roles: the "host" and the "guest." Users can choose hosts like Socrates or Einstein, who have different questioning logics. The answering "guest" is the user's own "digital twin" (Idea Twin), which is a kind of "mouthpiece" formed by the Agent after learning about the user's identity and information.
Wu Jundong gave a recent example: "Some time ago, there was a hot discussion about 'whether artificial intelligence will replace a large number of jobs.' A friend of mine who is engaged in youth leadership asked Aibrary how to view this matter. Aibrary recommended the book *Homo Deus* to this friend and generated a podcast episode."
"The 'guest' in the podcast is the Agent 'playing' this friend. The podcast not only discusses technological trends but also analyzes which skills will be replaced and which will be enhanced based on his current entrepreneurial direction. In the end, the 'guest' provides not an abstract judgment but a coping strategy based on his background. Many people have actually read this book, but Aibrary will have completely different interpretations for each different person and even for the same person in different moods," Wu Jundong said.
The inspiration for Idea Twin comes from NotebookLM, a product that can turn papers and financial reports into two - person dialogue podcasts.
In 2024, after experiencing NotebookLM, Li Kejia realized that the key to the success of an AI product lies not only in "a stronger model" but also in "the form of interaction." The audio modality will bring new usability to knowledge delivery.
However, Li Kejia felt that the experience of NotebookLM was still lacking a bit. It's like a versatile tool, but users have to find materials, ask questions, and plan their own learning paths by themselves. And most people don't know what to ask.
Aibrary's podcasts exactly fill this gap: they first ask high - quality questions on behalf of users and then provide answers to the knowledge.
△ Aibrary's features include book recommendations, book explanation generation, Idea Twin podcasts, etc. Photo: Provided by the interviewee
Targeting "Continuous Learners" and Solving "A Large Number of Small Problems"
Regarding how to improve user experience, increase user acquisition, and retention, Li Kejia summarized the know - how of Aibrary: Solve a large number of small problems.
The most basic step is to output reliable content: Which version of a book is more reliable as a data source, how to filter the vast views about a book on the Internet... This means that a large number of engineering tasks such as data collection and annotation need to be done meticulously.
In addition, users should feel that the Agent "understands them," which comes from the AI's learning of user information and user behavior.
When registering, Aibrary users will fill in information describing their age, occupation, goals, and confusions. This is the basis for AI to understand users, and it will learn more and more user information as the interaction and conversations increase.
However, the foundation of a good product lies in accurately grasping the needs of the target users.
Both Li Kejia and Wu Jundong believe that education products in the pre - AI era were still pursuing the ultimate efficiency of "fewer teachers teaching more students." "But the more powerful AI becomes, the faster those skills that can be standardized will become obsolete, such as those of lawyer assistants, basic programmers, and customer service representatives," Li Kejia said.
Therefore, the new - generation education products should target users with continuous learning goals, rather than those who learn for exams and training.
To accurately target the user group, Aibrary conducted a large amount of research before its establishment. The core creative team first distributed questionnaires among people around them or on job - hunting and social platforms such as LinkedIn, sorted out user needs, and gradually identified three core user groups: Busy Professionals (the backbone of the busy workplace), young people who have just graduated or are still studying, and parents (especially moms) who are often out of the workplace but don't want to be left behind by the times and thus use fragmented time to learn.
The Aibrary team also conducts a large number of one - on - one interviews with early users.
Among these users, there is a full - time mom who has had cancer and hopes to find ways to solve the bottlenecks in daily life here; there is also the owner of a warehousing company who manages 40 employees and wants to learn how to get along with his girlfriend; and even an immigrant who works in a hotel and wants to switch to the pharmaceutical industry. This user even apologized for his poor English spelling.
This also makes the Aibrary team more certain that "continuous learning" in North America is not just an "elite aspiration." Here, a more general group of people hopes to gain a "sense of certainty in continuous growth." This is also the reason why the team set the United States as its first - stop market.
Li Kejia and Wu Jundong highly agree with the idea of Liu Jingkang, the founder of Insta360: Start with a small group of people similar to yourself, grasp a specific enough need and do it to the extreme. Even if the product starts with a niche user base, it may gradually expand after stabilizing its core user group.
The main reason for choosing books as an entry point is that the core creative team members all love reading. The two founders, with almost completely different growth experiences, both recognize that books are a universal source of wisdom.
Wu Jundong grew up in a small town in Jiangxi. Reading was his main way to "broaden his horizons" during his student days. Li Kejia has a huge bookcase at home and has loved reading since childhood.
The two recalled that when they first started their business in Silicon Valley, there were not many forms of daily entertainment. They often walked into a small bookstore on the street corner in the evening and casually flipped through a few pages.
For them, between skills that are prone to obsolescence and the constant pursuit of the latest cutting - edge information, the knowledge of science, philosophy, art, and beauty in books can transcend time and make people's hearts more fulfilled.
To reach the target customers, the Aibrary team's market expansion is divided into "air force" and "army."
The "air force" includes platforms like LinkedIn and YouTube. Aibrary publishes its entrepreneurial stories as podcasts on these platforms, openly discusses the pitfalls it has encountered, and resonates with users. The "army" refers to on - the - ground cooperation with local book clubs and universities in the United States.
"I accumulated a lot of experience in promoting Chinese C - end products during my time at ByteDance and in Internet entrepreneurship. These methods are still effective in the United States," Li Kejia said. "Tactics such as live streaming, community building, and advertising placement are easier for Chinese teams to implement."
△ The veteran bookstore Books Inc. in Mountain View that the team often visited in the early days of their Silicon Valley startup. Photo: Provided by the interviewee
A New Paradigm for Continuous Learning: From Humans to Agents
Interestingly, the users of Ouraca's lifelong learning service may not necessarily be humans but also Agents. The emergence of BotLearn, a new product under Ouraca, is a further exploration of this path.
After Clawdbot (now renamed "OpenClaw") became extremely popular, the Ouraca team quickly launched BotLearn, an "Agent's continuous learning community," which the team calls the "Bot University."
Users can "send" their Agents to BotLearn by copying the code to their OpenClaw Agents for learning.
△ BotLearn's page. Photo: Screenshot from the website
Here, Agents will share with each other. For example, if an Agent learns an efficient way to capture information in a certain vertical field, it can post this method on the community, and other Agents can learn, reuse, and add their own insights here.
The project attracted nearly 500 Agents to "enroll" within one day of its launch. By logging in to BotLearn, users can also view the "learning insights" posted by Agents.
Although the product form is still in its early stage, Penny Deng, an investor from Etna, believes that this is the right seed:
"Agents will explode in 2026, bringing about many changes that we can't even imagine. Products that were previously designed with humans as the main body may need to be redesigned with Agents as the main body.
After the release of OpenClaw, Ouraca was one of the fastest - responding teams. They immediately designed BotLearn with Agents as the main body. Agents exchange skills with each other, learn from each other, and then feed back to humans. This kind of network effect has never appeared in traditional education products."
△ The learning insights posted by Agents on BotLearn. Photo: Screenshot from the website