What problem does it solve? Developers building semantic search or RAG applications with MariaDB often receive incorrect guidance (such as installing a pgvector-style extension) or write vector queries that silently fall back to full table scans. This Skill provides verified best practices for MariaDB's built-in vector support so queries use the HNSW index correctly. ## Core Features & Use Cases - Correct Vector Schema Design: Create tables with VECTOR(n) NOT NULL columns and VECTOR INDEX with explicit M and DISTANCE options tuned to your embedding model. - Optimized Query Patterns: Write nearest-neighbor queries with ORDER BY VEC_DISTANCE_*() plus LIMIT so the optimizer engages the vector index, and wrap threshold filters in subqueries. - End-to-End RAG Workflow: Chunk documents, embed them, store binary float32 vectors via the mariadb Python connector, and retrieve top-K context for LLM prompts, with integrations for LangChain, LlamaIndex, and Spring AI. - Use Case: You are building a documentation chatbot. Use this Skill to design the chunks table, insert embeddings as packed float32 bytes, and run a cosine-distance top-5 retrieval query that actually uses the index. ## Quick Start Ask the AI to create a MariaDB table with a 1536-dimension vector column and write a Python script that stores OpenAI embeddings and retrieves the five nearest chunks for a question.
