Turn an invisible pipeline into something inspectable
Retrieval-augmented generation is often described as “giving AI your documents.” That skips the decisions a developer needs to understand: which passages are selected, what reaches the prompt, and what the answer can actually support.
The visual RAG course exposes these stages in a fixed example. The evidence inspector on this site's homepage narrows the idea further: a return-policy question has support, a delivery question does not, and removing the policy removes the basis for either answer.
Separate teaching a mechanism from running a model
The demonstrations run in the browser with authored examples. They do not call a language model, calculate real embeddings or query a vector database. That choice lets the learner inspect a stable cause-and-effect relationship without needing an account, a key or a paid service.
It also sets a boundary. These demos cannot measure retrieval quality, hallucination rates, latency or production reliability. A live system needs evaluation against its own documents, tasks and failure cases.
Keep the learning surface usable
The site uses a Python static builder with HTML, CSS and JavaScript. Reading content is rendered into pages; browser interactions add quizzes, visual explanations, flashcards and local progress. Courses and reference pages remain separately addressable, so a useful explanation can be linked directly.
Local progress keeps the learning flow lightweight. It belongs to that browser: it is not an account, a cross-device sync service or proof that a learner has mastered the subject.
What to inspect
- The RAG course: follow the retrieval and context flow.
- AI Basics: explanations and checks for an early learner.
- The learning lab: questions, search and flashcards.
- The editorial policy: how explanations are drafted and checked.
The engineering principle
A polished answer can hide weak evidence. An interface can do the opposite: keep the source close, preserve its conditions, and make an unsupported answer obvious. That is the principle this learning product makes tangible.