~/projects/llm-post-classification $ cat README.md
Classifying 550 social posts with an LLM
Using an LLM to sort around 550 social posts by what readers need from them.
- n8n
- JavaScript
- Claude Haiku
- Anthropic API
// the problem
The team wanted to see which posts serve which reader needs. Labelling hundreds of posts by hand against a framework is slow, and people get inconsistent.
// what I built
A workflow in n8n and JavaScript that sends each post to Claude Haiku together with the team's reader-needs framework and gets a structured label back. Around 550 posts went through it.
// how it works
The framework goes in with every post, so the model labels against the same definitions every time.
┌───────────────────────────────┐
│ Social posts │
└───────────────┬───────────────┘
↓
n8n / JavaScript
↓
Claude Haiku ← reader-needs framework
↓
Classifications for analysis// result
A labelled set of around 550 posts the team could analyse. The data and findings stay internal.
// what I took away
An LLM is good at applying a clear framework to hundreds of posts. Figuring out what the results mean was still up to the team.