Retrieval (RAG)

What you'll learn

1. 1

   Embeddings & Similarity

   Turn meaning into numbers so you can search by meaning, not just keywords.

   • Meaning as coordinates
   • Explore an embedding space
   • Implement cosine similarity
   • Reading scores

   Take-home template: Semantic Search in 40 Lines

2. 2

   Chunking Documents

   Split documents into retrievable pieces, without cutting the meaning apart.

   • Why split documents?
   • Tune chunk size and overlap
   • Implement chunking with overlap
   • Chunk size trade-offs

   Take-home template: Recursive Text Chunker

3. 3

   Building a RAG Pipeline

   Connect indexing, retrieval and generation into one working system.

   • Two phases
   • Watch: a whole RAG system in 30 lines
   • The query-time flow
   • Retrieve the top-k chunks
   • Build the grounded prompt
   • Find the hallucination guard

   Take-home template: Production RAG on Supabase pgvector

4. 4

   Hybrid Search & Reranking

   Combine keyword and vector search, then rerank for precision.

   • Vectors aren't enough on their own
   • Implement Reciprocal Rank Fusion
   • Which search wins?
   • Measuring retrieval

   Take-home template: Hybrid Search: BM25 + Vectors + RRF

5. 5

   Evaluating RAG: Recall, MRR & Faithfulness

   Measure retrieval and generation separately, so you know exactly what to fix.

   • Two halves, two scorecards
   • Recall@k
   • Mean Reciprocal Rank
   • Retrieval or generation?
   • Diagnose it

   Take-home template: RAG Retrieval Eval Script