Writing
Notes from building and running things.
Practical notes, guides and lessons from building and operating real products.
3 articles
Featured2026 · 09 · 137 min read
What Vector Embeddings Are, and Where They Help
What vector embeddings are, the four product problems they solve, where a plain index still wins, and why pgvector beats a vector database first.
AI
·10 min readCut a 300s Product Filter Query to 30s With a Lookup TableA correlated EXISTS join across a 12M-row product/category pivot took 300s. Denormalizing the filter columns into one indexed lookup table dropped it to 30s.Performance
·9 min readThe 15-Filter Query That Broke Our MySQL SearchOur audio search hit 15+ filters across full-text, ranges, and facets, and no MySQL index could cover them. Why faceted search needs a search engine.Architecture