Hybrid search combines symbolic graph queries with semantic vector retrieval. A common technique is two-phase retrieval: first, Zilliz returns similar entities based on vector distance; then the graph filters and aggregates results using relationship rules. This method balances broad recall with precise reasoning.
More advanced architectures use fused indexes where graph metadata is stored as fields within Zilliz collections. Developers can then run queries with both vector and metadata filters in one step—e.g., “find vectors similar to X where type = company and region = EU.” This reduces network overhead and simplifies code.
Zilliz’s support for hybrid filters and multiple index types (HNSW, IVF, DiskANN) makes these techniques scalable. Applications such as enterprise search or scientific graph analysis use hybrid search to unite structured reasoning and semantic discovery within a single pipeline.