I wanted to learn sentence transformers and cosine similarity hands-on, so I built a news aggregator as a practice project that runs every hour and works with an editorial layer that runs on Telegram.

It scrapes AI news, clusters semantically similar stories together, and serves a single static HTML page — updated daily. Deliberately minimal UI (Craigslist-inspired).

Stack: Python, sentence-transformers, static HTML/CSS, GoatCounter, DigitalOcean VPS.

Still tuning the clustering. Next up: story labels, summaries, and an editorial layer.