🛠 Playground

Learn by doing. These exercises run entirely in your browser with deterministic mock models — drag, tweak and experiment freely. No setup, nothing to break.

Exercise 1 · Build the RAG pipeline

The five stages of a RAG system are shuffled below. Drag them (or use the ▲▼ buttons) into the order they actually run, then check your answer.

  1. 1Generate grounded answerThe model answers from the context.
  2. 2Retrieve top-k chunksFind the most similar documents.
  3. 3Embed the queryTurn the question into a vector.
  4. 4User questionThe query that starts everything.
  5. 5Augment the promptInsert retrieved context into the prompt.

Drag the cards (or use ▲▼) so data flows top-to-bottom in the order a RAG system runs.

Exercise 2 · Tune the retriever

Retrieval is a balancing act. Adjust top-k and the similarity threshold and watch which documents make it into the context — too few and the model starves, too many and it drowns in noise.

Retrieving 0 of 0 documents — nothing clears your filters; the model would have no context!

    Want the theory behind these? Read the runnable RAG project and the 18 RAG Techniques lab.