I’m quite new to the n8n platform and to this community, so please bear with me if this is a beginner’s mistake — still getting familiar with the flow logic and node interactions.
I’ve set up an n8n environment running locally on an Ubuntu 22.04 LTS machine, using Docker & Portainer. The machine has an Nvidia GPU, and I’ve successfully installed the main tools involved: Ollama, Qdrant, and n8n.
What I’m trying to build:
The flow is designed to process a transcript (usually an interview with some patient information), and I want to query this transcript via a set of predefined questions — each with its own custom prompt.
I’ve used the following structure in n8n:
chatTrigger to handle incoming questions.
if statement to check if a file was uploaded or if it’s a new question.
If a new file is detected, it gets loaded and split.
Then Ollama creates embeddings, which are stored in Qdrant.
Finally, the AI Agent is supposed to search the VectorStore and generate an answer using only the stored embeddings.
The problem:
Although the flow seems to work technically (no errors, the flow runs end-to-end), I’m not getting the expected retrieved results.It looks like the VectorStore is being queried, but it doesn’t return any useful matches or content.
I also wonder: How would you go about iterating over a list of predefined questions with their own prompts (instead of manual input)?
In the end I found the solution, apperantly the node has a flaw to ollama, so I changed the node on the RAG AI agent to the openai one, and then set the API-key to anything and set the Base URL to the local ollama instance.
How would you go about iterating over a list of predefined questions with their own prompts (instead of manual input)?
My Flow: