RAG Playground

Embed documents, then watch a query retrieve the nearest chunks.

Open the RAG Architecture lesson →

1. Indexing the corpus (offline, done once)

  1. Documents → Chunks
  2. Embeddings
  3. Vector DB

2. Answering a query (online, every question)

  1. User Query
  2. Embed Query
  3. Similarity Search
  4. Top-k Chunks
  5. LLM
  6. Grounded Answer

Source documents

Pick up to 3; each adds to the text.

ChunkOverlapNext chunk22 chunks are outlined in the text; text shared by two chunks has a yellow grid background.

3,016 / 8,000 chars

Strategy

Group whole sentences up to the target size.

200 chars
30 chars

22 chunks

Runs all-MiniLM-L6-v2 in your browser. The first embed downloads about 23 MB of model weights.

Embed the source documents to plot every chunk here.