AI Data: Why Vector Databases Aren't Enough скачать в хорошем качестве

AI Data: Why Vector Databases Aren't Enough

Why SQL, designed over 50 years ago, is here to stay! In this episode of Tanzu Talk, Ivan Novick and ‪@thecote‬ discuss how AI applications actually connect to enterprise data, with SQL. They talk about: Why vector databases are only part of the solution (they're for unstructured data like images and video, not your transactional data) How AI applications use SQL to efficiently query existing databases The role of metadata and data catalogs in helping AI understand what data is available Why 50 years of SQL optimization still matters for AI performance and cost (the difference between 10¢ and 1¢ per query at scale) Real-world examples using shipping ports and industrial farms How Model Context Protocol (MCP) tools connect AI to databases The importance of data architecture and pipelines in AI applications Lessons learned from the NoSQL movement that apply to AI today AI needs efficient access to your existing structured data, and SQL remains the most optimized way to provide that. Vector databases handle unstructured data, but SQL databases handle everything else, and a lot of that is probably your enterprise data. Ivan's blog post: https://thenewstack.io/why-ai-and-sql... Learn more about Tanzu Data products: https://tanzu.vmware.com/data Index: 00:00 Tanzu Talk - Ivan on AI and SQL 01:21 Vector Databases 03:40 Vector database example: drone footage at a port 05:27 Vector database example: farms 06:56 Optimizing with data architecture 08:48 Connecting data to the AI with tools 09:52 Optimizing data for the AI 14:53 What do the MCP tools do? Finding and understanding the metadata 19:45 How metadata is used for optimized queries 25:55 Getting data optimized for AIs - the enterprise data pipeline 31:35 Optimization example: iceberg 33:13 Lessons from NoSQL