Vector databases cheat sheet

1. 1Embeddings and similarityWhat a vector is in this world, where it comes from and the three ways of measuring how close two vectors are.8 concepts

   1. 1EmbeddingA fixed-length list of floating-point numbers that a model produces for a piece of text, an image or other data, placed so that similar meanings land close…
   2. 2Choosing an embedding modelThe model that maps input to vectors.
   3. 3DimensionsThe number of values in each vector, fixed by the model, commonly between 384 and 3072.
   4. 4Cosine similarityThe cosine of the angle between two vectors, from -1 (opposite) to 1 (same direction).
   5. 5Dot productThe sum of the element-wise products of two vectors.
   6. 6Euclidean distanceThe straight-line distance between two points: the square root of the summed squared differences.
   7. 7NormalisationScaling a vector to length 1 by dividing it by its norm.
   8. 8Semantic vs lexical searchSemantic search compares meaning through embeddings.

   8 in this chapterOpen this chapter
2. 2Nearest neighbour searchFinding the k closest vectors to a query, exactly or approximately, and why approximate is usually the right call.6 concepts

   1. 1Exact k-nearest neighboursCompare the query with every stored vector and keep the k closest.
   2. 2Approximate nearest neighboursSearch with an index that visits only a small part of the data and returns most, but not always all, of the true nearest neighbours.
   3. 3Top-k and similarity thresholdsk is how many neighbours a query returns.
   4. 4Curse of dimensionalityIn high dimensions, distances between random points bunch together and classic space-partitioning structures such as k-d trees stop pruning much.
   5. 5Recall of an ANN indexThe share of the true top k neighbours, found by exact search, that the ANN search also returned. 0.95 means 95 of every 100 true neighbours were found.
   6. 6When exact search is enoughBelow roughly tens of thousands of vectors, or after a filter narrows the candidates to a few thousand, a full scan is often fast enough.

   6 in this chapterOpen this chapter
3. Index types: HNSW, IVF and PQThe main ANN index families, what each one trades and the knobs that matter.8 conceptsOpens with credits
4. Recall, latency and filteringTuning an index against real numbers, and combining vector search with ordinary conditions such as tenant, language or date.6 conceptsOpens with credits
5. Hybrid search and rerankingCombining keyword and vector results, and using a stronger model to reorder the shortlist.7 conceptsOpens with credits
6. Chunking for RAGRetrieval-augmented generation depends on what you store. How you cut documents into chunks decides what the model can ever be shown.7 conceptsOpens with credits
7. pgvector in practiceVector search inside PostgreSQL: the column types, the distance operators and how to query it from Python.7 conceptsOpens with credits
8. Dedicated stores and choosingThe main purpose-built vector databases, what sets each apart, and how to decide between them and Postgres.7 conceptsOpens with credits
9. Evaluation and operationsMeasuring retrieval quality, and running vectors over time as models, data and costs change.8 conceptsOpens with credits
10. 10Glossary, A to ZEvery concept in the sheet and the smaller terms it uses, in one list.88 terms

Embeddings, similarity metrics, ANN indexes, filtering, hybrid search, reranking, RAG chunking, pgvector, dedicated stores and evaluation.