Microsoft · AI-200
PostgreSQL, vector search and RAG
AI-200 · Lesson 7 of 13 · 46:38
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the AI-200 exam (Microsoft Certified: Azure AI Cloud Developer Associate). In this lesson: pgvector and its distance operators, HNSW, IVFFlat, and DiskANN indexes, operator classes and query-time tuning, RAG queries with metadata filters and thresholds, memory sizing, and PgBouncer in transaction mode. Includes three original practice questions.
Preview the topics
- A table has an HNSW index built with vector_l2_ops. Queries sort by cosine distance, and run as slow sequential scans. What is wrong?
- A support bot must only retrieve chunks from the user’s product line, and it must say that it does not know, rather than use weak matches. Which query elements fit?
- Through PgBouncer in transaction mode, an app runs SET hnsw.ef_search = 100 once, right after connecting. Later searches still behave as if it were 40. What should the app do?