AI · Interactive learning

RAG Playground

An interactive observatory for retrieval-augmented generation, with the vector space drawn at each step. Explore embeddings, hybrid retrieval, reranking, and evaluation directly in the browser.

How do you understand a system you cannot see?

Retrieval-augmented generation brings several moving parts together. Documents become chunks, chunks become vectors, and retrieval decides which context a model gets to see. A final answer alone hides most of those decisions.

RAG Playground makes those intermediate steps the subject of the experience. The vector space is visualised along the way, so retrieval becomes something to inspect rather than an opaque step behind a chat box.

The models download to the visitor’s machine, and the documents stay in the browser. Embeddings, hybrid retrieval, reranking, and evaluation run locally. That brings the learning experience and the privacy boundary into the same place.

Keep exploring

Follow the connected note for a closer look at the ideas behind this work.

Make retrieval visible before making it clever

Find a thread

Search across projects, notes, and ideas.