Tracking Vietnamese news trends with embeddings, HDBSCAN and an LLM

Major events are typically covered by many outlets under different headlines, which makes it hard to see the full picture from individual articles. A developer built a system called Tin tức hôm nay có gì? that reads about 95 Vietnamese RSS feeds, groups articles on the same subject into topics, and tracks how each topic's momentum changes over time. The pipeline runs in stages: ingesting and deduplicating articles, embedding their text into vectors, clustering them with HDBSCAN, updating each topic's lifecycle, and using an LLM to label and explain it. The service is written in TypeScript on Bun, stores its data in SQLite, and runs more often during peak news hours. The author chose HDBSCAN because it finds the number of clusters automatically and leaves unrelated one-off stories unassigned, which suits noisy news data.