AI News Curation Agent
Built a personal AI curation system that filters RSS feeds, scores articles against my interests, and delivers a daily digest via Telegram.
I was drowning in newsletters and RSS feeds. The interesting stuff was buried under mountains of content I didn’t care about, and I kept missing articles I actually wanted to read. So I built something to fix it.
How It Works
The system pulls from my RSS feeds (AI/tech newsletters, Hacker News, a few aggregators) and runs everything through Firecrawl to grab the full content. All of that lands in a SQLite database as a temporary holding area.
Then an LLM evaluates each piece against a profile of my interests. What survives the filter gets formatted and sent to me via Telegram every morning. If something’s worth keeping long-term, I can trigger a separate workflow to archive it to my Neo4j knowledge base.
The Technical Bits
I went with SQLite for the temporary storage because the data is inherently disposable. Articles get evaluated, delivered, and cleaned up automatically. The schema tracks articles, curation results, and my feedback so the system can learn what I actually click on.
Keeping the processing database separate from my permanent knowledge base was a deliberate choice. I didn’t want years of “maybe interesting” articles bloating a system meant for things I’ve actually read and found valuable.
The whole pipeline runs in n8n and typically finishes in under 5 minutes. Daily delivery has been rock solid, and the relevance ratio stays above 80%, which means most mornings I’m reading useful stuff instead of deleting noise.