| RAG: Retrieval, Embeddings and Evaluation |
Why models need retrieval; embeddings and vector similarity; chunking strategies; semantic, keyword and hybrid search, reranking and query routing; building the RAG prompt with citations; query transformation and parent-child chunking; retrieval and generation metrics and building a test set |
Open |
| Lab 1: RAG on a Real Company Handbook |
Build an assistant over 38 pages of GitLab's public handbook that answers with sources and says "I don't know" when the documents don't cover a question; compare four chunking strategies and see what metadata filters change. |
Notebook Colab |
| Lab 2: Evaluate Your RAG |
Turn a spot check into an evaluation: build a test set, measure keyword, vector and hybrid retrieval with hit rate and MRR, check answers for correctness and faithfulness, check the LLM judge against your own labels, and report error bars. |
Notebook Colab |
