Skip to content
← All casesUnsupervised ML + EvalAI

Real News — anti-doomscroll aggregator

HDBSCAN clustering of ~20 sources → LLM rewrite → automated bias-check before publishing on a fixed schedule

At a glance
HDBSCAN — density clustering with no preset k

Founder, architect & solo developer

solo · solo

  • Python
  • HDBSCAN
  • DeepSeek
  • Telegram
  • RSS

Expand the deep-dive ↓

Problem

What doesn't work

News feeds are infinite and anxiety-inducing; one story is duplicated by dozens of sources; manual selection and dedup don't scale.

Solution

Architectural approach

Unsupervised ML + eval: HDBSCAN groups ~20 sources into stories (no preset cluster count), an LLM rewrites each story neutrally, a separate bias-check pass verifies tone before release. Fixed windows instead of an infinite feed.

My role & contribution

Founder, architect & solo developer

Built the aggregator: collection from ~20 sources (10 TG + 10 RSS), HDBSCAN story clustering, an LLM rewrite of each item, an automated bias-check pass before publishing. 3 publishing windows a day.

Ready to discuss?

If you need an architect who builds autonomous AI systems — reach out.

Serbia-based · CET/CEST timezone · EU-aligned working hours · International contracts experience