aipath: An Interactive, Browser-Based AI Course with 30 Lessons and No Math Prerequisites

What aipath is and who it is designed for

aipath is an interactive AI general education course built as a Vite and React web application. It targets absolute beginners, defined in the README as people who want to understand AI concepts without working through mathematical derivations. The README explicitly states zero math as the course design principle: it uses visualizations and interactive demos to build intuition before introducing formal terminology.

The live version runs at aipath.buynao.com with no signup required. The full application is also available as a self-hostable repository for organizations that want to run it internally or adapt the content.

The course covers the arc from foundational AI concepts to practical implementation. The README describes the target transformation as taking someone from an AI-news bystander to an AI-app builder over 30 lessons of about 20 minutes each. Each lesson follows a fixed structure: core concept with intuitive explanation, an interactive demo where the learner adjusts parameters directly, common pitfalls to watch out for, and a mini exercise. A fixed table of contents on the left of each lesson page allows jumping to any lesson at any point.

How the six stages and 30 lessons are organized

The curriculum divides into six stages, each with a distinct focus.

Stage 1 covers intuition: what AI is, how machines learn, how a single neuron works, gradient descent, and the relationship between data and overfitting (lessons 1 through 5).

Stage 2 covers the four pillars of deep learning: backpropagation, convolutional neural networks, embeddings, attention mechanisms, and the Transformer architecture (lessons 6 through 10). The README describes the attention mechanism section as "highlighting what matters" rather than starting with the self-attention formula, and diffusion as "wiping a picture out of noise."

Stage 3 covers large language models: tokenization, pretraining, supervised fine-tuning and RLHF, temperature and sampling, and scaling laws (lessons 11 through 15).

Stage 4 covers practical application: prompt engineering, context management, retrieval-augmented generation (RAG), function calling, and agents (lessons 16 through 20).

Stage 5 covers the frontier: diffusion models, multimodal models, reasoning models, the MCP ecosystem, and the current LLM landscape (lessons 21 through 25).

Stage 6 is the hands-on coding stage: calling an LLM API directly, local model deployment with Ollama, building a RAG knowledge base, evaluation and safety red lines, and a final lesson on where to go next (lessons 26 through 30).

Running aipath locally and the development setup

The repository uses Vite with React 18. To run a local instance:

npm install
npm run dev

The development server starts with hot reload at http://localhost:5173. To build for production:

npm run build

This outputs to the dist/ directory. To preview the production build locally:

npm run preview

The preview server starts at http://localhost:4173. The package.json lists React 18.3.1, React Router DOM 6.28.0, Recharts 2.13.3, and Three.js 0.160.1 as runtime dependencies. Three.js is used for the 3D word vector visualization described in the README as a starfield of word vectors you can navigate in 3D. The build is a Vite SPA, so the dist/ output is a set of static files that can be served from any static hosting service.

The README notes that the entire codebase, including the design system and React architecture, was generated by Claude Fable and verified lesson-by-lesson in a headless browser. Complex interactions using Three.js, Canvas, and declarative SVG were specifically noted as requiring headless verification.

How the interactive demos work

Each lesson's interactive component is the primary learning mechanism. The README describes several examples. In the temperature and sampling lesson (lesson 15), a learner can turn a temperature knob and watch the model's top-five token candidates reshape in real time, from focused to more variable. In the gradient descent lesson, a learner drags a neuron's weights and sees the error surface respond. In the word embedding lesson, a learner roams a 3D starfield of word vectors.

The README states that the course was designed around the principle that what you have played with is what truly becomes yours, contrasting with reading a static explanation.

The RAG lesson (lesson 18) and the RAG in practice lesson (lesson 28) both let learners chop a document into chunks and feed them to a RAG system themselves. Stage 6 lessons are described as real code: calling an LLM API, running an open-source model on a local machine with Ollama, building a knowledge base, and reviewing evaluation and safety considerations before deployment.

The application's data architecture is described in the README as data-driven and auto-registered: the lesson catalog in src/data/lessons.js is the single source of truth for all 30 lessons, and the route table in App.jsx registers routes automatically from it.

Limitations and what the course does not cover

aipath is a conceptual course, not a technical reference. It explicitly avoids mathematical derivations throughout. A learner who finishes all 30 lessons will have built intuition and run some code examples, but will not have worked through the derivation of backpropagation, the mathematics of attention, or the formal theory behind sampling algorithms. The README frames this as a design choice: build the mental model first.

The course has no formal assessment infrastructure. There are mini exercises at the end of each lesson, but no grading, no certificates, and no instructor or community channel built into the repository itself. The README notes a Discord channel and a WeChat QR code for community, but these are external.

For learners who already have a Python background and want to go deeper, Fast.ai is a widely known alternative that teaches deep learning from code-first principles using PyTorch, working through real implementations rather than browser demos. Fast.ai is not browser-based: it requires Python, Jupyter notebooks, and GPU access for training exercises. aipath is better suited to true beginners who need to build a conceptual foundation first; Fast.ai is better suited to learners who are ready to train models.

The package.json lists the project as private, and no license is stated in the README or package.json for the course content or source code. Using or distributing the content without clarification from the authors carries licensing uncertainty.

Maintenance, language support, and licensing

The last push was on 2026-07-02, and there are no GitHub releases. The README's status section shows all 30 lessons migrated to Vite and React and passing the production build. Code splitting, dark mode, and the fixed left table of contents are listed as completed. The bilingual implementation covers all 30 lesson bodies, demos, and quizzes, with a built-in language switch that flips the entire site without a page reload.

The README notes the project was generated by Claude Fable, meaning the course content, copy, interactive visualizations, and React architecture were produced by an AI system and then polished. The README does not describe an ongoing roadmap for new lessons or content additions beyond the existing 30.

The package.json has no license field, and the README does not state a content license. The top-level license situation is unclear. Anyone who wants to fork, translate, or redistribute the content for educational purposes should seek clarification from the repository owner before proceeding.

Editorial conclusion

Absolute beginners who want a structured introduction to AI concepts before writing any code will find aipath's staged layout and interactive demos genuinely useful: the progression from intuition through large model internals to hands-on API calls is coherent, and every concept has a playable demo rather than a static diagram. The course is not suited to readers who already have a working understanding of transformers or who need formal assessment with certificates; there is no grading, no instructor feedback, and no community forum built into the repository. The lack of a stated license is a practical barrier for anyone who wants to adapt or redistribute the content: the package.json marks the project as private and the README does not specify a content license. The last recorded push was on 2026-07-02.

Frequently asked questions

Official sources

1. buynao/aipath on GitHub
2. Issues
3. Project website
4. README