Classify LinkedIn posts as quality or slop with OpenAI and Qdrant

Description

Purpose

This workflow is the official backend for the StopSlopIn Chrome extension – it classifies LinkedIn posts as quality or slop using a strict LLM quality gate and learns from user votes over time via a Qdrant vector store.

What this is for

This runs the webhook that powers the StopSlopIn Chrome extension on the Chrome Web Store. The extension sends LinkedIn posts here for analysis and user votes here for training – everything stays on your own n8n instance.

Setup

Add your OpenAI credentials to the chat model and embeddings nodes
Add your Qdrant credentials to both vector store nodes, pointing to a collection named stopslopin
Activate the workflow, copy the webhook URL, and paste it into the StopSlopIn Chrome extension settings
Follow the instructions on the yellow sticky notes for anything else

How it works

A single webhook exposes two actions, selected via a ?action= query parameter: analyze for classification, vote for training.

A Switch node routes incoming requests based on the action parameter
On analyze: each post is enriched with similar prior-rated posts pulled from Qdrant (RAG), batched together, and sent to the LLM with a strict quality-gate system prompt
The LLM returns a structured JSON array of pass / fail results, which is sent back to the caller
On vote: the post is embedded and stored in Qdrant along with the user's "good" or "slop" rating as metadata
Every new vote becomes in-context reference material for future classifications, so the filter gradually adapts to personal taste

Customization

LLM** – swap the OpenAI Chat Model for any LangChain-compatible chat model (Claude, Ollama, etc.)
Prompt** – edit the system prompt inside the Basic LLM Chain node to match your own feed taste
Similarity threshold** – change the value in the Filter node (default 0.7) to make RAG examples looser or stricter

Compatibility

n8n Version 2.17.0 or above
Cloud or Self-Hosted
Requires: OpenAI account, Qdrant instance

Note: post contents sent through this workflow are forwarded to the configured LLM and embeddings provider (OpenAI by default). Swap those nodes for a local or alternative provider if that is a concern.