Home / How it works
How it works guides
17 published guides. Start with How semantic search works.
Encode, index, retrieve, rank: the full semantic search pipeline with what each stage decides and how it fails.
Read the guideLearning path
All guides
Understand it3 guides
Start hereHow semantic search worksEncode, index, retrieve, rank: the full semantic search pipeline with what each stage decides and how it fails.ExplainerDense vectorsWhat a dense vector is, why every dimension carries meaning, and how dense vectors power semantic matching.ExplainerSparse vectorsSparse vectors store weights for a few terms out of a huge vocabulary. How they work and why they still matter for search.
How it works12 guides
ExplainerSemantic search architectureA reference architecture for production semantic search, with the indexing path and the query path drawn separately.ExplainerVector searchVector search explained: how queries and documents become vectors and how the nearest ones are found.ExplainerSimilarity searchSimilarity search finds the items closest to a query in vector space. How it works and how it differs from exact lookup.ExplainerSimilarity metrics: cosine, dot product, EuclideanWhich distance metric to use for semantic search, how each is calculated, and when normalisation makes them equivalent.ExplainerApproximate nearest neighbor searchHow ANN search trades a little recall for large speed gains, and which index families semantic search systems use.ExplainerSemantic indexing and vector indexesWhat a vector index stores, how it is built and updated, and why index choice decides speed and recall.ExplainerTop-k retrievalWhat k controls, how it interacts with filters and reranking, and how to pick it from data rather than habit.ExplainerRelevance scoresWhat a similarity score means, why raw scores are not comparable across queries, and how thresholds go wrong.ExplainerSynonyms and related termsWhy embeddings match synonyms without a synonym list, and where they still confuse related but different terms.ExplainerMultilingual and cross-lingual semantic searchHow one embedding space lets a query in one language find documents in another, and how to evaluate it.ExplainerSemantic matchingSemantic matching scores how close two pieces of text are in meaning. Methods, models and common failure cases.ExplainerChunking for semantic searchHow to split long documents so each chunk is retrievable on its own, and how chunk size changes recall.