A Conversation with Chunxiang Zhao: How a Liberal Arts Major Built a Top-3 AI App

By | Muxin Xu

Edited by | Jing Liu

In any Lanzhou hand-pulled noodle shop during the '90s, when a customer ordered "large bowl, second-thin, no cilantro," the apprentice cooks in the back would echo the order back in ringing Lanzhou dialect, one after another. The head chef would explain the reasons to his trainees: first, so the customer feels heard; second, to project energy and spirit; third, it's a rule passed down from the ancestors — you shout every day you pull noodles.

In a sense, this was the most primitive form of interaction. User input enters through the order, output echoes from the kitchen. Zhao Chunxiang's childhood was filled with these shouts. Later he became a screenwriter, weaving them into his scripts. When he pivoted to indie development, this same logic became foundational to how he built AI applications.

Zhao's "Stomach Book" (Wei Zhi Shu) is a large language model-based food journaling app that leverages LLM summarization rather than generation. Users snap photos to log daily meals; the AI assists with calorie counts, metabolic pathways, and dietary analysis. Zhao has three principles for building AI apps: first, it must be beautiful; second, deliver the most spectacular payoff in the shortest possible time; third, design primarily for women.

As a freemium app with in-app purchases, Stomach Book hit $12,000 in revenue during its launch month of May, climbing to #3 on Apple's Food & Drink chart (#1 was Xiachufang).

Among today's AI entrepreneurs, Zhao is an outlier. He's not from Tsinghua, Peking, Fudan, or Jiaotong. He's not a technical specialist in any field. A chemistry major turned signed author at Han Han's company, he became a full-time screenwriter after graduation. Then he switched to product management, entered the startup world during mobile internet's "latecomer" phase, and burned through one million yuan in six months. To cope with the pain of failure, he studied Chinese philosophy and taught courses — building a base of users that would later support his pivot to coding and app development.

He peppers technical discussions with humanistic asides: "The structure of Crime and Punishment is as rigorous as code," or "If I make a billion, I'm dropping everything to make films."

In the current AI wave, Zhao's story is simple: it's about a slash-generation kid who stumbled his way into the AI era.

The following is a conversation between An Yong Waves and Zhao Chunxiang —

An Yong: You've had strong showings at several AI demo days lately, landing some TS on the spot. What won over the investor judges?

Zhao Chunxiang: Because I finally explained the secret of Transformer to them. Products driven by Transformer architecture have a fundamentally unsound business logic, because lagging models can always extract answers from advanced models — and those answers are themselves the advanced model's trade secrets. When an advanced model's outputs become training data for lagging models, no commercial moat can be created.

An Yong: How did you reach this conclusion?

Zhao Chunxiang: From Google's original Transformer paper. And this isn't a conclusion — it's a fact. The industry calls it "distillation," which is really just taking your answers to train my own AI, dressed up as "extracting the essence." Haven't you noticed AI responses getting more and more "AI-flavored"? From this fact, I offered three corollaries:

1. The LLM price war will never end, meaning the field will never see a "tripartite balance" where only a few leaders dominate a space, like Douyin and Kuaishou, or Huya and Douyu, or Meituan and Dianping.
2. Due to the Transformer problem outlined above, first-mover AI apps cannot build commercial moats.
3. To escape these two paradoxes of commercial moats, AI app entrepreneurs must build community moats and brand mindshare instead.

An Yong: So for AI app entrepreneurs, how should they actually build their products?

Zhao Chunxiang: I have "three don'ts." One: don't rely on LLM generation capabilities — by the way, I think AI companionship born from this is a false need. Two: don't do multi-function or general-function; go vertical. Three: don't let your app be strongly associated with AI in public perception, because ultimately you're competing with traditional apps like Douyin and Xiaohongshu for user time. Users won't set aside dedicated time for you just because you're an AI product.

An Yong: So the investors voted and offered TS because they bought your reasoning?

Zhao Chunxiang: Yes, plus the app was trending on Xiaohongshu. Another factor: the model companies' rounds are definitely already oversubscribed, so my theory was comforting to those who missed out.

An Yong: What happened with those TS afterward?

Zhao Chunxiang: Nothing. Investors got an adrenaline spike on the spot, then came down.

An Yong: You've been hitting so many founder events lately — were you fundraising for Stomach Book?

Zhao Chunxiang: Yes. But after talking to everyone, I realized Stomach Book itself may not be fundable. Its ceiling is too low. The top Food & Drink app is Xiachufang, with monthly revenue of $120,000 — just 12x Stomach Book's. VCs won't back projects with such low growth multiples.

An Yong: Sounds like your investor conversations didn't go smoothly?

Zhao Chunxiang: Investors would rather give money to founders with better pedigrees. Because even if this one fails, at least I've paid for experience, so I can invest in them a second or third time. When they see I used to write novels, their hearts probably sink.

An Yong: What do they ask you?

Zhao Chunxiang: They ask "what's your original motivation for starting a company," and whatever I say, they respond: "Wrong. Actually you want money." I definitely need money, but personally, beyond a certain point it's useless. Money is discourse power and leadership — you can do bigger things with it. We don't really love money; we just love having more than others.

An Yong: What are your costs now?

Zhao Chunxiang: LLMs are in a price war, and I'm the only developer. Basically zero cost.

An Yong: You've built many apps, but Stomach Book seems to be your first LLM-integrated product?

Zhao Chunxiang: Because Stomach Book's need is also an old need — similar to solo dining logs on Xiaohongshu. But where Mint Health takes 15 steps to log a meal, Stomach Book takes three. So this is innovation in functional experience enabled by AI, but fundamentally I don't think this wave of AI apps can create new needs, the way short video did in the mobile internet era.

An Yong: You don't seem to care much about AI technology — that's where you differ from many AI app entrepreneurs.

Zhao Chunxiang: Often AI "gods" will build an app around a technical innovation. I think that's making dumplings just for the vinegar, but users don't care how cutting-edge your vinegar is. To me, AI is just a wrench in the toolbox — not that high-status. I won't build a product around AI. I stay closer to the market, closer to user perception.

An Yong: Has this AI wave excited you?

Zhao Chunxiang: It's like making a deal with the devil: I can make AI very human-like, even pass the Turing test, but you'll never gain wisdom from it. Generative AI has sucked all the oxygen from the room — AI development had multiple paths, and now the others have vanished. I think generative AI only pursues correct answers, not wisdom. If humanity goes all-in on this single path, I'm pessimistic. Language is an abstraction of the world, and AI's summarization and vectorization of language is abstraction upon abstraction. There's not more information — there's less.

An Yong: So how did you go from screenwriter to AI app developer?

Zhao Chunxiang: I started writing novels in my sophomore year, became a signed author for Han Han's One project, making 10,000 yuan a month, 8,888 after tax. So I stopped going to class, nearly failed. My original goal in writing was filmmaking. As a student I couldn't enter the film industry yet, so novels were a "budget alternative" — a novel is a film one person can shoot. After graduation, I naturally joined Han Han's company to write scripts. But gradually I realized the path from junior screenwriter to film director was too long; I needed money. It was 2019. I chose to enter tech, becoming a product manager. Product management was ten times easier than screenwriting. What fascinated me: you create a product, then distribute it to millions of phones through an update package — like distributing my novels, but to a far bigger audience.

An Yong: What products did you build?

Zhao Chunxiang: I did an offline social product called Space Chat. It deployed to 200 venues in Shanghai; anyone who paid by phone could join the same group — an app that clustered people with similar spending power and aesthetics. Then a Shanghai family-office investor gave me one million yuan.

An Yong: What happened to it?

Zhao Chunxiang: By 2019, the mobile internet story was already over, but I thought the future was still bright. So I burned through one million yuan in six months and completely failed. The failure was devastating. I lay in my Shanghai rental with the lights off, feeling abandoned by the world. In that process I encountered Chinese philosophy — it made me feel personal failure is just a pebble misplaced in the grand sweep of history. I made videos on Chinese philosophy for various platforms, totaling 50 million views. I ran two philosophy courses with over 100 students. Then I got back on the road. I taught myself to code, kept building products — a digital memorial flower, a to-do list app called Zhuan Shan, various utilities. Until Stomach Book, my most successful product so far. Stomach Book is a handsome bow tie on my ragged, hard-worn shirt — and that bow tie depends on every past failure.

An Yong: How hard was teaching yourself to code as a humanities student?

Zhao Chunxiang: Not too bad. I discovered code is like classic literature — Crime and Punishment, for example — with a beautiful, barely perceptible thread connecting every character and line. Code has that same precision.

An Yong: Did LLMs make building AI apps easier?

Zhao Chunxiang: Not really. AI apps used to require heavy backend development; APIs simplified that. The hard part is frontend. Almost all AI app frontends are chatboxes now, and users are sick of them.

An Yong: How much did your PM experience help with AI app development?

Zhao Chunxiang: Even with APIs, being able to build apps myself rests on knowing traditional internet frontend, backend, and mobile — AI uses streaming too now. But my screenwriting and film experience is also indispensable. It gave me an aesthetic sensibility. Given equal capability, users will always choose the more beautiful app.

An Yong: What about screenwriting specifically?

Zhao Chunxiang: My life always has cinematic moments. When founding companies, I'd write ideas on walls like in A Beautiful Mind. And I've always thought: if I ever make a billion, I'm dropping all code to make films.

An Yong: How do you make a billion?

Zhao Chunxiang: Keep writing code. Leave the rest to fate.

Image source | Stomach Book app

Layout | Yao Nan