The AI engine for super individuals — code × media dual-core infrastructure.
Turn your expertise into keynotes, pitch decks, and full courses.
What • Pipeline • Who • Use Cases • Architecture • Roadmap • FAQ • 中文
Herline (赫能) is a vertical AI engine built for the 12M+ solopreneurs, independent consultants, and knowledge creators who have deep expertise but struggle to turn it into external deliverables.
Unlike generic AI tools (ChatGPT, Claude, Notion AI) that help you write words, Herline runs a 7-layer pipeline that turns your professional knowledge into the three forms of outbound expression every super individual actually uses — speeches to deliver on stage, pitch decks to raise capital, and courses to teach paying students.
🌐 Live at herline.vip — free tier available.
Smart people read 50 books a year and still can't teach what they learned. Not because the books are bad — but because from "I read it" to "I can teach it", an entire workflow is missing:
- Deconstruct — knowledge networks broken into reusable blocks, not just summaries
- Recompose — blocks reorganized by learning arc, not outline stacking
- Deliver — expressed in the right medium (a keynote ≠ a course ≠ a short video)
- Persist — saved as a personal knowledge graph, not forgotten after the chat closes
Generic AI tools break down at step one. Herline is the pipeline that closes the loop.
The mission: make every knowledge worker a viable one-person company.
"Code and media are permissionless leverage." — Naval Ravikant, 2018
Naval split wealth leverage into Labor / Capital / Code / Media. The latter two are permissionless — you don't need a boss's approval to write code or publish work. Herline operationalizes this observation into a workflow: a code × media dual-core infrastructure on the company side, growing cognitive-depth × public-influence dual-axis capability on the user side.
For most of the last century, "good teaching" was scarce because four ingredients took years to compound: deep knowledge, cross-domain composition, error-pattern intuition, and practice reps. Generative AI is equalizing all four at the speed of a software release — so professional scarcity migrates to areas AI cannot equalize as fast.
We're betting on two axes:
- Cognitive depth — a defensible point of view in a domain, built through the loop of using AI to interrogate, structure, and pressure-test what you already know.
- Public influence — a multi-year archive of work others have followed, which compounds slowly and cannot be retroactively manufactured.
Hiring signals are shifting. Alongside credentials, employers are increasingly reading public archives — code, posts, recorded talks, courses you've shipped — and how you collaborate with AI in real workflows. This pattern is showing up earliest in AI-native roles, but the signal generalizes outward.
Herline is built to grow both axes in the same workflow:
- Every deep-read fills the personal knowledge graph → depth.
- Every Prep export becomes a public artifact → influence.
- The pipeline runs as a daily compounding loop, not a one-shot generator.
This is also why generic "AI productivity" tools and stand-alone "speaking" programs each address only half of the picture. The two are best developed together.
📖 Longer version with falsification conditions: English docs/two-axis-thesis.md · 中文 docs/two-axis-thesis.zh-CN.md.
┌─────────────┐
│ Assessment │ → Capture your professional profile
└──────┬──────┘
↓
┌─────────────┐
│ Atlas │ → Personalized book & content planning
└──────┬──────┘
↓
┌─────────────┐
│ Library │ → Deconstruct books into reusable knowledge blocks
│ (D2B) │ (Claims / Concepts / Methods)
└──────┬──────┘
↓
┌─────────────┐
│ Studio │ → Design course strategy from your knowledge graph
└──────┬──────┘
↓
┌─────────────┐
│ Courses │ → Auto-generate full courses with scripts,
│ (B2C) │ TTS audio, and assessments
└──────┬──────┘
↓
┌─────────────┐
│ Prep │ → Export keynotes, pitch decks, teaching materials
└──────┬──────┘
↓
┌─────────────┐
│ Boost │ → Adapt content for multi-platform distribution
└─────────────┘
Each layer is usable standalone. Or chain them end-to-end: one book in, a complete set of deliverables out.
📖 See docs/architecture.md for a deeper walkthrough.
| Persona | Core Pain | What Herline Solves |
|---|---|---|
| Independent Consultant | 10 years of expertise reassembled from scratch for every proposal | Personal knowledge graph → reusable callable blocks |
| Content Creator / Lecturer | 2–4 weeks to prepare a single course | D2B → Studio → B2C pipeline compresses it to 1–2 days |
| Solo Founder | Pitch decks eat creative energy before product work | Expertise + one book → pitch deck ready for iteration |
| Institution / School | Need 10 curricula/year, can only produce 2 | One teacher's expertise auto-expands into a curriculum matrix |
- 🧠 Persistent personal knowledge graph — grows with every book, every deep-read, every course. Cross-book RAG surfaces insights at intersections you'd never find manually.
- 🏗️ End-to-end vertical pipeline — not another AI chat box. Seven specialized layers, each addressing a specific workflow break.
- 🌐 Bilingual native — Chinese and English are first-class throughout, not bolted-on translation layers.
- 📦 Export to deliverables, not documents — PDFs with timing notes, PPTX templates with speaker structure, curriculum-ready assessments.
- 🎯 Built for super individuals — 12M+ one-person companies, domain experts, educators. Not for teams, not for enterprises (yet).
- 🔐 Data ownership — your knowledge database belongs to you. The system gets smarter with use; we never repackage your content. See
docs/data-handling.md.
Input: a book you've read + a short bio. Output: a 30–45 min keynote script with timing annotations, slide deck template, and quote cards. Typical time: ~30 minutes end-to-end.
Input: your professional profile + target audience. Output: 15–20 page pitch deck with narrative arc and speaker notes. Typical time: ~45 minutes end-to-end.
Input: a book + target learner profile. Output: 3–7 unit course with scripts, TTS audio narration, and assessments. Typical time: 1–2 hours end-to-end.
Every read, course, and deck persists in your personal knowledge database. Cross-book RAG surfaces connections you'd never find manually. The longer you use Herline, the smarter it gets for you specifically.
At a high level:
| Layer | Stack |
|---|---|
| Frontend | React 18 · Next.js 16 · TypeScript · Tailwind CSS |
| Backend | FastAPI · SQLAlchemy (async) · Python 3.11 |
| AI orchestration | LangGraph 1.0 — multi-step workflows across specialized agents |
| Data | MySQL 8 (metadata) · Redis 7 (cache/queue) · Milvus 2.6 (vector) |
| Async processing | ARQ on Redis Streams — long-running D2B / B2C generation jobs |
| Speech synthesis | Cloud-based TTS for audio course generation |
Under the hood:
- LangGraph-orchestrated multi-step pipelines with state checkpointing
- Dual-LLM strategy: generative models for synthesis, reasoning models for structured output
- Cross-book RAG via Milvus with custom embedding strategies
- Token-aware cost management across 10M+ monthly token operations
- Structured knowledge-block schema (Claims / Concepts / Methods) that enables cross-book composition
The deconstruction and generation engines are proprietary. This repository exists for public documentation, not source distribution. See docs/architecture.md for a deeper walkthrough.
- 7-layer pipeline — production-ready
- Prep export: PDF / PPTX / DOCX
- Cross-book RAG over personal knowledge graph
- Elective courses marketplace
- Sharing & collaboration primitives
- GCYSC — Global Chinese Youth Speaking Tour, international school program (overseas-only; out of scope for domestic K–12 admissions)
- City partner program V1 for the AI-literacy curriculum
- Marketing agent (Boost) for automated multi-platform distribution
- Sentiment & citation tracking for brand visibility in AI search
- MCP Server — native Claude / ChatGPT / Cursor integration
- Public OpenAPI spec — enable external tool integrations
- GPT Store presence for "Course Planner by Herline"
- International expansion (EN / ES / JA)
- Creator monetization layer
No. This repository is public documentation and community resources only. The engines that power Herline's deconstruction and generation are proprietary — they represent years of domain-specific research and are what differentiate Herline from generic AI wrappers. We may open specific peripheral components (like MCP integration and OpenAPI spec) over time.
Generic AI tools are one-shot. You ask, they respond, and the context evaporates. Herline is a pipeline with persistent state — your personal knowledge database grows with every use. Long-term Herline users become significantly more productive than ad-hoc AI users because of this compounding effect.
Yes. The pipeline is bilingual throughout. You can deep-read English books, generate English courses, and export English-language keynotes. Native support, not translation layers.
Yes. The institution tier supports curriculum matrices and teacher collaboration. One teacher's expertise auto-expands into a full curriculum matrix. Contact [email protected] for institutional inquiries.
Your knowledge database belongs to you. We never repackage your content, never train models on your private data, and never sell data. You can export your knowledge graph at any time. See docs/data-handling.md.
- D2B deep read: 8–15 minutes per book (Claims / Concepts / Methods extraction)
- B2C course generation: 30–90 minutes for a 3–7 unit course with audio
- Prep export: under 2 minutes per format
- Long runs execute asynchronously — you can queue multiple jobs.
Yes. Sign up at herline.vip — the free tier covers the assessment, first deep-read, and a course generation trial.
See herline.vip/privacy for the full privacy policy. Enterprise compliance features (SSO, audit logs, data residency) are available on higher tiers — contact [email protected].
More questions? See docs/faq.md for the extended FAQ.
- 🌐 Product — herline.vip
- 📖 Documentation —
docs/(architecture · glossary · FAQ · data handling) - 🏛️ Architecture walkthrough —
docs/architecture.md - 📚 Glossary —
docs/glossary.md - 🎓 Example outputs —
examples/ - 📋 Awesome Super Individual — curated resources for solopreneurs (companion list, CC0)
- 💬 Discussions — GitHub Discussions
- 📧 Contact — [email protected]
- 🤝 Partner / institutional inquiries — [email protected]
We welcome translations, documentation improvements, and case study contributions. See CONTRIBUTING.md.
Core product feature requests go to [email protected] or herline.vip/feedback — not via this repo.
Herline (赫能) is built by a small team operating out of East Asia and beyond, shipping production infrastructure for a global audience of super individuals.
We believe the defining challenge of the next decade isn't "can AI generate content" — it's can every knowledge worker turn their expertise into products that compound. Herline is our answer.
Mission: make every knowledge worker a viable one-person company — by closing the workflow gap between what they know and what they can deliver.
The content in this repository (documentation, diagrams, examples, and resources) is licensed under CC BY 4.0. You are free to share and adapt, with attribution.
Herline's core engine is proprietary and not part of this repository. See herline.vip/terms for product terms of service.