Vector Search Integration

Embedding and Searching Support Knowledge with Qdrant

Overview

To make the AI agent truly helpful, it needs access to real company knowledge: FAQs, policies, SOPs, product documentation, and more.

Instead of hard-coding these responses, this system uses Qdrant (or Pinecone) as a vector database. This allows the AI to search your support docs semantically and respond with high-confidence, accurate answers — even if the user asks the question in a new way.

1. What You Store in the Vector DB

Each entry in your knowledge base (FAQ, doc, template) should include:

Field	Description
`question`	The original FAQ or common query
`answer`	The ideal support response
`category`	e.g. refunds, onboarding, billing, etc.
`source`	(Optional) Link to source doc or help page
`last_updated`	For tracking freshness of info

2. Embedding the Knowledge

To embed your knowledge base:

Use OpenAI’s text-embedding-3-small model
For each FAQ or doc:
- Send the question field to OpenAI’s embedding API
- Store the resulting vector + metadata in Qdrant

n8n Steps:

Add a Function Node to format your questions
Call OpenAI embedding API
Insert into Qdrant using their Upsert API

3. Querying During Support Flow

When a message is classified (e.g. “faq” or “how_to”):