Added Semantic Categorization Via Embeddings To Articles/Threads On My Website

You can now filter my articles by topic at https://illya.sh/threads/topics/

Qwen3-Embedding-0.6B is used to vectorize and embed all of the articles/threads, which are then tagged/categorized based on natural language tags for the category.
This allows me to define a new category by defining a new Python class with the topic/category name, description, URL path, and the query sting to embed and query against the thread embeddings. From there, all of the content is generated statically.

Everything runs 100% locally/offline. Model weights are pre-downloaded once - that's the most time-consuming part.

The motivation for implementing this has to with the content's volume. I have written over 180 articles/threads, but until now there was no categorization. So if you wanted to read an article on a particular topic, you'd either have to google it, ask an LLM or go through the list manually yourself. Now, there is an index with topics, which help navigating the content semantically.

Navigate to https://t.co/bXb1JWKD4d, select a topic, and you will be presented paginated list of all threads/articles matching that topic.