Rockset achieves billion-scale similarity search, driving scalable AI apps
Rockset, the search and analytics company, is unveiling expanded vector search capabilities that achieve billion-scale similarity search in the cloud. Approximate nearest neighbor (ANN) search, the latest capability, builds off Rockset’s investment into supporting machine learning (ML) and AI applications.
With generative AI’s (GenAI’s) explosive popularity in the past year, growing costs threaten to impede its advancement. Large language models (LLMs), which create vector embeddings with thousands of dimensions, make exact nearest neighbor search extremely expensive and complex.
Support for ANN by Rockset greatly increases the company’s ability to power voice assistants, chatbots, anomaly detection, recommendation and personalization engines, and more, at scale, according to the company. By storing and indexing billions of vectors—as well as hundreds of terabytes of metadata, including text, JSON, geo-, and time-series data—enterprises leveraging Rockset will be able to produce relevant AI experiences at the scale in which its application demands.
Rockset’s expanded vector search functions, when used in tandem with LlamaIndex and LangChain integrations, can further accelerate AI development.
“Enterprises will only continue to leverage AI if they have the ability to scale AI applications efficiently, which is why Rockset is designed for billion-scale vector search in the cloud,” said Venkat Venkataramani, co-founder and CEO of Rockset. “Efficiently incorporating real-time signals and updates into vector search applications is no easy feat. We’ve spent years designing Rockset for real-time updates and are thrilled that companies can now build AI applications at scale.”
Additionally, Rockset’s search index, accompanied by an integrated SQL engine, enables metadata filtering that is as simple as SQL WHERE clauses, according to the company.
AI applications can also be built with real-time updates via inserting, updating, and detecting vector and metadata with indexes built on RocksDB. Any new data is represented in searches in milliseconds, without incurring any costly reindexing. For greater confidence in scaling AI apps in production, Rockset customers can separate indexing and search with compute-compute separation.
To learn more about Rockset’s expanded vector search capabilities, please visit https://rockset.com/.