Self-hosted AI news agent for n8n: RSS → LLM scoring → Postgres → Telegram

Built with n8n
1

Hi everyone,

I built a self-hosted n8n workflow bundle called NewsScoreAgent. It pulls articles from AI/tech RSS feeds, cleans and deduplicates them, scores them with an LLM, stores the best items in Postgres, and optionally sends a Telegram summary.

:backhand_index_pointing_right: Product page (screenshots, docs, ZIP download):

Main flow:

  • RSS feeds → normalize & deduplicate

  • LLM scoring (WOW score 1–10)

  • filter by threshold

  • Postgres UPSERT on article URL

  • optional Telegram daily summary

  • optional weekly TXT export from Postgres

Why I built it:

  • I wanted a simple self-hosted research pipeline for AI/robotics news.

  • I also wanted persistent storage in Postgres instead of just sending notifications.

  • The workflow is designed to continue gracefully if RSS items are missing, full-text fetch fails, or the LLM returns invalid JSON.

Tech details:

  • n8n 2.15.1

  • Postgres 13+

  • Groq by default, but the scoring step can be adapted to other OpenAI-compatible providers

  • URL-based deduplication via UPSERT on link

  • optional full-text fetch

  • daily summary + weekly export flow

I attached:

  • a workflow screenshot

  • a Telegram output screenshot

  • a small sample of the DB structure / schema

I’d love feedback on:

  1. workflow structure

  2. robustness / failure handling

  3. whether this kind of bundle would also make sense as a public template / marketplace item

Happy to answer any technical questions about the setup, scoring logic or Postgres part.

NewsScoreAgent

1 Like
5

That’s a very good idea — and I’ll test it.
Per‑feed calibration keeps each source in its optimal quality band.