AWS Tests Enterprise Search Service Backed by Machine Learning
Amazon Kendra, a new natural language search service for enterprises, is now available for public preview.
Amazon Web Services (AWS) announced the Amazon Kendra public preview earlier this month as part of its 2019 re:Invent conference. The service, which can run as a console application or as an API, lets users apply natural language queries to their companies' various content sources and receive "highly accurate" answers.
The service works across all kinds of information sources, allowing companies to "get rid of information silos," according to the Amazon Kendra product site. The public preview version comes with connectors for Microsoft SharePoint Online, Java Database Connectivity and Amazon S3. Other connectors to cloud-based services including Salesforce, Microsoft OneDrive, Dropbox and Box are expected to be added once the product becomes generally available (AWS has not said when that would be, however).
Other features in the public preview include support for keyword searches and natural language queries, the ability to retrieve results from unstructured data and FAQs, document ranking, relevance tuning, domain optimization and, as mentioned, a limited set of built-in connectors.
Features that will be added upon general availability are analytics, query auto-completion, incremental learning and more connectors.
AWS plans to offer Kendra in two flavors -- a Developer Edition for testing purposes and an Enterprise Edition for production use. Pricing information is available here. Currently, the public preview is available only in the Northern Virginia, Oregon and Ireland regions.
AWS' announcement of Amazon Kendra's closely follows similar news from its cloud rival Microsoft, which detailed its own efforts around AI-driven enterprise search at its November Ignite conference. Microsoft is currently running a private preview of a new "semantic search" capability for organizations.
Gladys Rama is the senior site producer for Redmondmag.com, RCPmag.com and MCPmag.com.