Drawbly

Drawbly · Your ideas, in your own lines

RAG pipeline

Separate preparing the knowledge from answering a question.

RAG pipeline with documents prepared into a searchable index above, and a question retrieving context for a model below.
Use this starter to explain a document-answering feature without squeezing every implementation detail into one diagram.

What this diagram shows

Preparation divides source documents into useful chunks and adds them to a searchable index. At question time, retrieval selects relevant content and the application supplies it with the question to a language model. The model generates an answer using that context. Vector search commonly adds embeddings; keyword and hybrid retrieval are also possible. Retrieved material can improve grounding, but does not guarantee a correct answer.

For the two paths and common failure points, read What is a RAG pipeline?

Make it fit your system

  1. Rename Documents and Search index for your own knowledge sources.
  2. Add your retrieval method: keyword, vector or hybrid search.
  3. Label the checks or citations your application uses before showing an answer.

Click Use this template to open the drawing. Choose Text to change a label, or Select to move a mark. If a drawing is already open, you can keep it or open the template with a recoverable copy.

Watch the explanation unfold

A captioned walkthrough, without narration. Open Video on the canvas to change the replay pacing or record your own explanation.

Technical reference: Microsoft: Retrieval-augmented generation.

The drawing and supplied marks are editable example content. No account or cloud saving is required. Free exports include Drawbly credit.

← All templates · Render a recipe locally