Why semantic-search quality depends on the model, language, and corpus — and how Dzen Chat lets you run a local model or switch providers.