Multi-vector search improves retrieval accuracy by using more than one semantic representation to find relevant information. A single query might be represented in multiple ways to capture different aspects of its meaning, and the system retrieves content that matches any of those representations.
It is particularly valuable for complex or ambiguous queries where a single vector might not fully capture what the user is looking for. Rather than hoping one representation covers everything, multi-vector search hedges across several, producing more complete and relevant retrieval results.
