RAG chatbot architecture diagram
This is a worked example. Wavelength built the diagram below from the description on this page. Nobody drew it by hand.
What was said
Help center articles are exported from Zendesk Guide. An ingestion job splits the articles into chunks, creates embeddings, and stores them in Pinecone. When a customer asks a question in the chat widget, the API retrieves the closest chunks from Pinecone and sends them with the question to the language model. If the model is not confident, the chat hands the customer to a support agent in Intercom.
The diagram
What is in this flow
- Systems: Zendesk Guide, Chat Widget, API, Language Model and Intercom.
- Data stores: Pinecone.
- Decisions: Model Confident?.
- Size: 10 steps and 9 connections.
Questions to confirm
The AI inferred these links. On a real call, you ask the customer to confirm them.
- Does Language Model check confidence Model Confident??
Mermaid code
Copy this into any Markdown file, GitHub README, or Mermaid Live to render the same flow.
flowchart TD
n1(["Help Center Articles")]
n2{{"Zendesk Guide"}}
n3["Ingestion Job"]
n4[("Pinecone")]
n5{{"Chat Widget"}}
n6{{"API"}}
n7{{"Language Model"}}
n8{{"Intercom"}}
n9{"Model Confident?"}
n10(["Customer Question")]
n1 -->|exported from| n2
n2 -->|ingests| n3
n3 -->|stores embeddings| n4
n10 -->|asks via| n5
n5 -->|sends question| n6
n6 -->|retrieves chunks| n4
n6 -->|sends chunks and question| n7
n7 -.->|check confidence| n9
n9 -->|No| n8Map your own process
Describe your version out loud, or paste the transcript of a call. Wavelength draws the flow as you talk, and you can edit it by voice, by prompt, or by hand.
Every new account starts with 14 days of Pro. No card needed.