Cloud search and analytics database Rockset claims it is paving the way for the next generation of vector databases, with the indexing, retrieval and ranking capabilities required for hybrid search.
The Rockset database has been upgraded to enable users to index any data – including vectors, text, document, geo, and time series – to create the most relevant results.
The rapid development of AI models – like Meta’s Llama-3, OpenAI’s GPT-4, Google’s Gemini, and Databricks’ DBRX, among others – is bringing in a new era of generalizable intelligence, according to Rockset, requiring “forward thinking” enterprises to invest in powerful retrieval systems built for AI applications.
Rockset promises to meet the demand, by introducing a new standard of vector database that encompasses rapid iteration on indexing, models and ranking algorithms.
Its database provides real-time indexing, full-featured SQL (structured query language) and cloud-native efficiency, to deliver “speed at scale,” said the provider. New features in the database update are ranking algorithms – including BM25 and reciprocal rank fusion – to build hybrid search applications, and multi-tenant design for retrieval augmented generation (RAG) applications.
Also, a new search design uses compressed bitmaps and covering indexes for faster performance at scale.
“All search will soon be hybrid search,” predicted Venkat Venkataramani, co-founder and CEO of Rockset. “Similarity search has limitations around domain awareness and requires different types of data – text, document, geo and time series data – to provide the necessary context.”
Venkataramani said support for hybrid search requires “best-in-class” indexing technology designed for fast retrieval. “Rockset is committed to continuously innovating our Converged Indexing technology, and we have now introduced text search and ranking algorithms for hybrid search.”
Converged Index enables users to build vector indexes without impacting live search applications, providing the flexibility to index any type of data and apply ranking and scoring using SQL. Such flexibility, claims Rockset, enables customers to build and iterate on search and AI applications faster to drive the most relevant experiences.
Last year, Rockset raised a further $37 million to aid growth with a B-round extension.
The company also expanded its vector search capabilities at the end of last year, with approximate nearest neighbor (ANN), achieving billion-scale similarity search in the cloud.
