I built a $3/yr AI workflow to stop doomscrolling Twitter for tech news

I’ve always had a simple but annoying problem: I want to stay updated on the latest open-source projects and tech trends, but I really don’t want to rely on scrolling Twitter to get them anymore. The signal-to-noise ratio is getting worse. To find a few genuinely valuable projects, most of my time is wasted on meaningless noise, outrage bait, and ads. I tried various aggregators and subscription feeds, but I still ended up having to click through and filter things manually. The core issue wasn't solved. So I built a small automated workflow called TechDistill. The goal is simple: Stop browsing the feeds entirely, and just get a curated result every day. Now, this pipeline generates four Markdown files daily: One overall Overview, to quickly catch up on what's worth paying attention to today. Three Source-specific reports (GitHub Trending, Hugging Face, Product Hunt) with more complete details. My daily routine looks like this: I scan the overview first to quickly judge if there's anything worth diving into. If there is, I check the specific source report for the details. Every project comes with a short blurb telling me exactly "what it actually does," so most of the time I don't even need to click into the README to understand it. The whole process is fully automated: scrape data directly from the sources → clean it up → let the model distill it. I basically don't visit the original sites anymore. It runs on GitHub Actions once a day, takes under 4 minutes, and costs ~$0.008 per run (OpenRouter + DeepSeek v3.2). I’ve been using it for a few days, and the most immediate change is this: I've actually stopped reflexively checking Twitter for tech news. For me, the point of this project isn’t "AI summarization." It’s about turning scattered, high-noise information into a stable, low-noise input stream that I can actually accumulate over time. Right now, it's mostly a personal tool, but I’m curious: is anyone else trying to solve this in a similar way, or do you have a better approach? GitHub: https://github.com/JunstinLee/TechDistill