Keywords vs Embeddings

Insights from building discovery feeds: from naive keyword extraction to naive embedder usage to better insights into how both work and might be combined to understand and match Bluesky posts(-ers). Based on your feedback I can make this more or less ATproto specific (e.g. scraping pitfalls), also more or less technical, leaning towards less technical. A light intro to TF-IDF (sparse) vs EmbeddingGemma (dense) vectors.