Case Study: Consensus achieves faster, more precise agentic literature search with Zilliz Cloud

Consensus boosts semantic search precision 14% with Zilliz Cloud

Consensus, a platform that provides an agent for scientific research, faced challenges scaling its search capabilities across over 400 million scholarly sources. Their existing keyword and sparse search methods struggled to find papers that used different terminology, and initial tests with dense-vector search in Elasticsearch resulted in poor quality, slow re-indexing times, and high storage costs. To power their move to an agentic architecture, which requires fast and precise retrieval, they needed a purpose-built vector database and chose to evaluate Zilliz Cloud.

Zilliz Cloud provided a managed vector database that powers the semantic search layer within Consensus's tri-brid retrieval system. The solution involved a daily full re-index of the entire collection in about an hour, enabling the use of larger, higher-quality vectors. The results included a 14% increase in search precision, P99 query latency of ~45 milliseconds, and up to 4x lower storage costs, which allowed for the deployment of larger vectors that further improved result quality by 27%.

Consensus