New Course: Vector Search in Practice

My new course Vector Search in Practice is now released through SQLskills.com.

This is the follow-up to Vector Search Essentials, about what happens when you run vector search in production. While the implementation is SQL Server 2025, I cover several topics that apply to vector search broadly.

Thirteen modules. Embedding model selection beyond MTEB leaderboard scores, including the cost tradeoffs between API and local inference. Schema design patterns for storing and versioning embeddings, with hash-based change detection so you know when things go stale.

I spend time on measuring search quality with precision@K, recall@K, MRR, and nDCG, and on where things can go wrong.

From there, reranking with Reciprocal Rank Fusion and cross-encoders, document chunking strategies including SQL Server’s AI_GENERATE_CHUNKS, and building RAG pipelines with SQL Server handling retrieval and a local LLM handling generation.

Everything runs locally with Ollama for the demos you can follow along with. No API keys required.

I really enjoyed working on this one - check it out: Vector Search in Practice (AIVSP)