Discussion with Sylvain, IT Support Lead
Sylvain manages a small IT support team. During a call, he asks me:
“Can we build a chatbot that only replies using our internal docs and nothing else?”
I reply:
“Yes. That’s RAG. Retrieval-Augmented Generation. Want a demo?”
To prove it, I create a test case using my own background as documentation.
I feed it into a local knowledge base, build the workflow in n8n using OpenAI, and connect it to a basic chat UI.
Then I start asking:
“Where did I work last year?”
“What tool do I use for automations?”
“Am I a designer?”
Each time, the bot answers with precision but only based on the file. When I ask something it doesn’t know, it replies:
“I don't have that information.”
Exactly as planned.
Problems addressed:
- Chatbots that hallucinate or invent info
- Support teams that need reliable assistants
- Lack of control over what the AI uses to reply
The solution I built :
A complete RAG pipeline using n8n :
- Google Drive : To store and download my file.
- Pinecone Vector Store: To store the document and make it vector.
Then in the same workflow :
- Chatbox : To ask a simple question
- Question and Answer Chain : To check into my vector
- Respond to Webhook : To send the answer into the chat
The result :
Sylvain tries it himself. He asks about our internal tools.
Accurate answers only. No guesses. No bullshit.
He says:
“Perfect. That’s the kind of assistant I’d trust.”
Mission done.