Hybrid Approach boosts Search Relevance by 45%
DataStax has unveiled a new capability aimed at enhancing AI-powered search systems, potentially transforming how businesses retrieve and deliver information to users.
The company's Astra DB Hybrid Search combines vector and lexical search methods to significantly improve search accuracy.
According to the company's announcement, the new technology improves search relevance by 45%, addressing a critical challenge for businesses implementing AI systems that require high levels of accuracy.
The solution integrates with NVIDIA's NeMo Retriever reranking microservices to automatically reorganize search results using fine-tuned large language models, resulting in more precise and contextually relevant information retrieval.
Ed Anuff, Chief Product Officer at DataStax, emphasized the importance of retrieval quality in RAG (retrieval-augmented generation) systems, noting that customers consider 95%+ accuracy "non-negotiable" for enterprise AI deployment.
The new capability will be available through Langflow, an open-source tool for low-code AI application development.
Developers can integrate the functionality using Astra DB Python client and schema-less Data API, with the system being hosted on Astra DB with GPUs for fast, cost-efficient processing.