Ainda não entendi exatamente como funciona. Mas esse artigo do Nick Hagar é mais um lembrete: vá entender o que é NLWeb e como funciona "Semantic Search".
> A Three-Step Pipeline. Behind the scenes, NLWeb operates on a straightforward three-stage pipeline designed for simplicity, cost-efficiency, and publisher control.
> Ingest: Standardized Item Collections. To cut down on complexity, it standardizes this process using schema.org conventions.
> Process: Efficient, Low-Cost LLM Logic. Next, when a user sends a query, NLWeb assesses relevance, scores results, and generates explanations.
> Serve: A Flexible, Headless API. Finally, NLWeb delivers its results via a flexible API. It is a “headless” system, meaning it has no built-in, mandatory UI.