Skip to content

CONCEPT Cited by 1 source

Semantic search retrieval

Definition

Semantic search retrieval is meaning-based search that retrieves related content based on conceptual similarity rather than exact keywords or explicit references. In contrast to keyword search (which requires the query and document to share literal terms) or reference-based lookup (which requires explicit links), semantic search can discover connections that exist conceptually but are invisible lexically.

Why it matters for system design

Many system design challenges involve connecting artifacts that describe the same thing in different vocabulary: - A threat model says "authenticate all API endpoints" while the implementing code uses @RequiresAuth decorators — no shared keywords, but semantically linked. - A design review discusses "rate limiting for the billing service" while the PR title says "add throttling middleware to payment-api."

Quantified impact (Dropbox)

At Dropbox, semantic search achieves what explicit references cannot: - 80% of design reviews linked to implementing code changes via semantic search. - Only 12% had explicit references (manual links). - 69% of connections were recoverable only through semantic search — invisible through any other mechanism.

(Source: sources/2026-06-12-dropbox-mcp-dash-design-to-code-security)

Relationship to hybrid retrieval

Semantic search is typically one component of a hybrid retrieval system. At Dropbox, the systems/dash-search-index combines BM25 lexical search with dense vectors — semantic retrieval handles the paraphrase/conceptual matches that BM25 misses.

Seen in

Last updated · 542 distilled / 1,571 read