Basics
| Item | Details |
|---|---|
| Founder | Zhao Chunxiang (former novelist, published 2 books, college dropout) |
| Products | BellyBook (胃之书) — AI diet tracking app; Stranger Alarm Clock; Wanla (AI bookmark tool) |
| Team size | 1 person |
| Revenue | BellyBook $12,000 in 5 months; #3 in Apple’s Food & Drink category; 90K users; received an investment Term Sheet (declined) |
| AI stack | LLM-driven diet analysis engine |
| Source | Woshipm / 36kr / Apptisan |
Storyline
Background: Zhao Chunxiang’s background is the most unusual in any of these cases — a novelist and screenwriter with 2 published books, college dropout, no technical or business training. AI took him from “content creator” to “product creator.”
Turning point: BellyBook’s core innovation is using an LLM for diet analysis — the user describes what they ate in natural language (“I had a bowl of Lanzhou noodles and one egg for lunch”) and the AI auto-parses nutrition and calories. It’s about 10x more convenient than manually searching a food database. Growth was entirely word-of-mouth, with zero marketing budget. Zhao charged for the product from day one, validating the principle that “users willing to pay are the only real users.”
Outcome: $12,000 in 5 months, 90K users, #3 in Apple’s Food & Drink category. His second product, “Stranger Alarm Clock,” got 8,000 signups in its first week. Zhou Hongyi (founder of 360) publicly recommended him. He received a Term Sheet from an investor but declined — he chose to stay a one-person company.
Workflow Breakdown
Minimum loop analysis
| Step | What he did | AI involvement | Notes |
|---|---|---|---|
| 1. Think it through | Found the “logging meals is too tedious” pain point from his personal weight-loss journey | 0% | The best product ideas come from your own real needs |
| 2. Write a spec | Defined features around user experience (“just say one sentence to log a meal”) | Medium | A novelist’s narrative thinking turned into product spec |
| 3. Implement | LLM API integration + AI-assisted coding | High | The core feature is LLM-driven natural language understanding |
| 4. Validate | Charged from day one and validated by payment rate | Low | ”Willing to pay = real demand” |
| 5. Iterate | Continuously improved AI understanding accuracy based on feedback | High | The AI’s diet-analysis accuracy directly drives retention |
3 Reproducible Key Moves
- Charge from day one: Don’t build a free product and “figure out monetization later.” Set a low-friction price ($1–$5/month) and validate with payment conversion, not signup volume.
- Use “novelist thinking” to build product: A product isn’t just a stack of features — it’s a “story.” The user is the protagonist; what problem does the product help the protagonist solve? Use this frame for App Store descriptions and marketing copy.
- Decline investment, stay one-person: Once revenue covers your living costs, ask yourself: “What do I need funding for?” If the answer is “to grow faster,” ask: “Faster for whom?”
Limits and risks
- Limited revenue scale: $12K in 5 months (~$2.4K/month) doesn’t yet fully cover living costs.
- AI accuracy is the core risk: If diet analysis is inaccurate, user trust collapses fast.
- Category ceiling: Diet tracking is a mature market (MyFitnessPal, etc.); differentiation requires continuous investment.
- Zero-marketing isn’t sustainable: Word-of-mouth has a ceiling — beyond a certain scale, active distribution becomes necessary.