DataStax has unveiled Astra DB Hybrid Search, which utilizes NVIDIA NeMo Retriever technology to enhance retrieval-augmented generation (RAG) systems. This capability reportedly improves search relevance by 45%. By integrating vector and lexical search, it aims to provide precise and contextually relevant AI-driven search results.

Ed Anuff, Chief Product Officer at DataStax, emphasized that accuracy in retrieval is crucial for enterprise Artificial Intelligence (AI) implementations. He noted, “We have heard from countless customers that attaining 95%+ accuracy is a non-negotiable when it comes to bringing enterprise AI into production. Astra DB Hybrid Search helps customers get there faster.”

This hybrid search method combines vector search, which focuses on semantic understanding, with lexical search, which ensures key terms are considered. The improved relevance aims to help users receive more accurate answers while utilizing generative AI applications.

Astra DB Hybrid Search also features NVIDIA NeMo Retriever's automation for text reranking, utilizing large language model (LLM) to reorder search results more effectively. This precision is intended to enhance user experiences in applications powered by AI.

For logistics software provider GoDash, the new capability is expected to streamline operations and provide insights for shipping customers. Aditya Swami, founder and CEO of GoDash, stated that the hybrid search will allow them to quickly retrieve relevant operational details, thus optimizing logistics processes.

Developers can implement this functionality using the Astra DB Python client and open API, allowing them to improve AI search and recommendations easily. The hybrid search is hosted on Astra DB, equipped with GPUs, providing efficient AI workloads without complex infrastructure.

The capability is also incorporated in Langflow, an open-source tool for low-code AI application development, allowing developers to rapidly experiment and refine search relevance.