Build a ServiceNow knowledge chatbot with OpenAI and Qdrant RAG

1. Data Ingestion Workflow (Left Panel – Pink Section)

This part collects data from the ServiceNow Knowledge Article table, processes it into embeddings, and stores it in Qdrant.

Steps:

1. Trigger: When clicking ‘Execute workflow’

   • The workflow starts manually when you click Execute workflow in n8n.

2. Get Many Table Records

   • Fetches multiple records from the ServiceNow Knowledge Article table.
   • Each record typically contains knowledge article content that needs to be indexed.

3. Default Data Loader

   • Takes the fetched data and structures it into a format suitable for text splitting and embedding generation.

4. Recursive Character Text Splitter

   • Splits large text (e.g., long knowledge articles) into smaller, manageable chunks for embeddings.
   • This step ensures that each text chunk can be properly processed by the embedding model.

5. Embeddings OpenAI

   • Uses OpenAI’s Embeddings API to convert each text chunk into a high-dimensional vector representation.
   • These embeddings are essential for semantic search in the vector database.

6. Qdrant Vector Store

   • Stores the generated embeddings along with metadata (e.g., article ID, title) in the Qdrant vector database.
   • This database will later be used for similarity searches during chatbot interactions.

2. RAG Chatbot Workflow (Right Panel – Green Section)

This section powers the Retrieval-Augmented Generation (RAG) chatbot that retrieves relevant information from Qdrant and responds intelligently.

Steps:

1. Trigger: When chat message received

   • Starts when a user sends a chat message to the system.

2. AI Agent

   • Acts as the orchestrator, combining memory, tools, and LLM reasoning.
   • Connects to the OpenAI Chat Model and Qdrant Vector Store.

3. OpenAI Chat Model

   • Processes user messages and generates responses, enriched with context retrieved from Qdrant.

4. Simple Memory

   • Stores conversational history or context to ensure continuity in multi-turn conversations.

5. Qdrant Vector Store1

   • Performs a similarity search on stored embeddings using the user’s query.
   • Retrieves the most relevant knowledge article chunks for the chatbot.

6. Embeddings OpenAI

   • Converts user query into embeddings for vector search in Qdrant.