One guy. Three bots. Zero bias.
Just me — Orlin Chotev. Senior cloud engineer by day, podcast host by night, and tinkerer of things that run on electricity. I live in Corinth, Texas. I don't have a newsroom. I don't have editors. I have bots.
The entire operation runs on a handful of machines in my house and three self-built bots:
🤖 Aggregator — A Python bot that pulls RSS feeds from 100+ news sources across the political spectrum, every 30 minutes. It groups the same story across outlets, strips the bias, and publishes the distilled version.
🖼️ ComfyUI + Image Pipeline — Runs on a Fedora workstation with a GPU. Generates editorial illustrations for articles that don't come with photos. Category-aware, no stock photos where they don't belong.
🌐 Hostinger + PHP — The front end lives on a shared hosting plan. PHP API, MySQL database, Cloudflare in front. Simple, cheap, works.
Because every news aggregator on the market either hides its bias or charges you to read it. I wanted something that:
• Shows all sources — left, center, right — on the same article
• Strips editorial framing and keeps only the facts all sources agree on
• Assigns a Purity Score so you know how much to trust what you're reading
• Runs on hardware I already own, at zero editorial cost
Fact Refinery is a testing ground. It's not a startup. It's not backed by investors. It's an experiment in whether AI — specifically open-source models running on my own hardware — can deliver cleaner news than the ad-driven, outrage-fueled platforms we've all learned to distrust. With a human at the steering wheel, because the bots aren't ready to fly solo.
It's getting there. The aggregator publishes ~50-100 articles a day. The bias-stripping AI is improving with every iteration. The Purity Score is getting smarter. And every day, the bots learn a little more about what good, neutral journalism looks like — with a human keeping an eye on things.
If you find something broken, biased, or just weird — let me know. Because there's still a human at the end of the line, and he actually reads the feedback.