TIL: truncating note bodies to 300 characters and front-loading #embeddings with structured metadata (title, tags, wiki-links) pushed my #obsidian vault recommender’s recall up significantly. The frontmatter I was already including in each note turned out to be the highest-signal input for the embedding model.

https://anoliphantneverforgets.com/til/2026-03-24-shorter-embedding-body-improves-recall