Curated example · Editable connections

RAG architecture
you can explore.

Separate document ingestion from retrieval and answer generation.

Curated example

Follow the connections.

Select a component to explore

Documents

Source material with ownership, freshness, and access rules.

Make it your own

Documents -> Chunking -> Document Embeddings -> Vector Store
Question -> Query Embedding -> Retrieval
Vector Store -> Retrieval -> Context Assembly -> Language Model -> Answer
Question -> Context Assembly

Download connections

Checking workspace availability. You can explore and copy this example now.

Two paths, one shared index

The ingestion path prepares documents for retrieval. The serving path embeds a question, retrieves relevant passages, and passes context to a model. Showing the paths separately makes it easier to discuss updates, latency, and where data is allowed to move.

Extend the RAG pipeline diagram

Add a reranker between retrieval and context assembly if your system uses one. Add an evaluation step outside the main serving path to inspect retrieval quality and answer grounding. Model names and vector databases are implementation choices; the responsibilities remain useful to show.

Make boundaries explicit

Identify which documents a user may access before assembling context. Mark stale or deleted sources and decide how the index is refreshed. This is a curated design example, not a deployed RAG application or evidence that a model will answer correctly.

Further reading: Google Cloud: RAG reference architecture ↗. The example above is independently authored and simplified.

RAG architecture
you can explore.

Follow the connections.

Make it your own

Two paths, one shared index

Extend the RAG pipeline diagram

Make boundaries explicit

Another system to explore

SaaS architecture

AI agent architecture