MindsDB explainer: SQL filters and hybrid search

MindsDB's November 24, 2025 upload introduces an open source federated query engine for searching unstructured data through SQL. The speaker argues that semantic search alone can return similar-looking items that miss a shopper's requirements. The example asks for an elegant iPhone 17 Pro Max case that is not black.

The overview briefly names Docker and built-in data source integrations, without giving a complete installation walkthrough. The speaker says federated queries run against data in place, then describes loading data into a knowledge base for semantic search. MindsDB handles chunking, embeddings, ranking and metadata matching; users can also supply their own vector stores, models and tuning.

The product search adds metadata filters to exclude black cases and require low return rates and high feedback ratings. An enhanced SQL-like operator searches review text for the meaning of "elegant," followed by a relevance threshold from the reranker. One unwanted match remains in the demonstration. Hybrid search adds keyword indexing, and an alpha parameter controls its balance with semantic similarity. The speaker reports more accurate results for this example, without establishing a general benchmark.

The final section describes a built-in AI agent that reads schemas, generates source-native SQL and performs joins. The speaker also names API, MCP and agent-to-agent connections for embedding MindsDB in an application.