Embeddings & Vector Databases · TRACK 02/04
Similarity Search & Indexing
Cosine vs dot product, ANN, HNSW — the math and indexes behind fast retrieval.
// THE TRACK
01 · START HERESemantic SearchLearn how search engines find 'laptop' when you typed 'notebook computer' — and how to build the same thing with embeddings.BEGINNERANN: HNSW & IVFUnderstand how HNSW and IVF let vector databases search millions of embeddings in milliseconds by trading a tiny bit of exactness for enormous speed.INTERMEDIATEHNSW IndexWalk through the layered graph structure that powers Pinecone, Qdrant, pgvector, and most other vector search engines.INTERMEDIATECosine vs Dot vs EuclideanGet an intuitive, visual grip on the three ways to measure vector closeness and a simple rule for choosing between them.BEGINNERIVF IndexLearn how an IVF index partitions vectors into clusters so search only scans the few buckets near your query, and how to tune nlist and nprobe.INTERMEDIATEProduct QuantizationUnderstand how product quantization shrinks embeddings by 10-30x by replacing vector chunks with codebook IDs, and what accuracy you trade for the memory.ADVANCEDRecall vs PrecisionLearn what recall@k and precision actually measure in vector search and how to benchmark your ANN index against ground-truth exact search.INTERMEDIATE