Hybrid Search: Combine Lexical and Vector Rankings
Combine lexical and vector candidates with deliberate ranking. Explore reciprocal rank fusion, candidate windows, duplication, and segment-level tests.
VECTORTOKEN LAB / TAG
Semantic Search articles
Retrieve meaning without losing the exact task.
Semantic search uses representations to connect differently worded queries and passages. Useful results still need to satisfy a concrete task: answering a configuration question, locating an appropriate procedure, or identifying a genuinely equivalent request. Topical resemblance is not enough when the passage omits the required instruction.
Use the retrieval guide to define an inspectable collection and an exact baseline. Read the similarity article to understand score direction and normalization without turning a decimal into a probability. Follow with hybrid search when exact names, version strings, or error codes expose gaps that another candidate signal might address.
Build evaluation queries with direct answers, paraphrases, similar but incorrect procedures, and no supported answer. Judge the expected behavior before inspecting model output. Keep source revisions and authorization boundaries attached to results. A good experiment explains both where the semantic signal helps and which important distinctions it still fails to preserve.
THE READING PATH
03 articles to explore.
Cosine Similarity in Vector AI: Geometry, Not Confidence
Work through cosine similarity, normalization, dot products, distance, and edge cases. Learn why a similarity score is not a confidence probability.
Token Vector Search: A Practical Retrieval Guide
Build an inspectable semantic retrieval baseline. Evaluate compatible embeddings, exact search, relevance labels, permissions, and candidate coverage.