Post by Pranav Bansal

Switched blog-feed's search from LIKE queries to SQLite FTS5.     - The old approach was WHERE title LIKE ? OR keywords LIKE ?. It worked for exact substrings but missed "deployment" when you searched "deploy," had no relevance ranking, and no snippet support. - FTS5 gives us proper tokenization and stemming out of the box, so prefix matching ("deploy" → matches "deployment") just works by appending * to the last token in the query. - Ranking uses BM25 via FTS5's built-in rank, weighted against the existing LLM quality score: (combined_score * 100) - rank DESC.    If the FTS5 index is missing for whatever reason, it falls back to LIKE automatically. Live at: https://lnkd.in/gRAV7nr2 Do let me know what you think. Also any contributions will be welcomed as well. #opensource #engineering #blogs #search #sideproject Blog Feed — Search engineering writing