Splice Machine 3.1 enhances support for real-time AI deployments

Splice Machine, a startup that offers offline and batch analysis tools to power intelligent apps for operational workflows, today launched version 3.1 of its platform. Splice Machine 3.1 introduces new features and functionality to support enterprises with real-time AI projects, including resource elasticity support on Kubernetes, GPU support, and extensions to Spark’s machine learning libraries.

Most companies struggle to develop working AI strategies. According to a recent survey by Rackspace, only 20% of enterprises report having mature AI and machine learning initiatives. Indeed, while Deloitte says 62% of respondents to its corporate October 2018 report deployed some form of AI, roughly 25% of companies see half their AI projects ultimately fail.

Splice says that version 3.1 of its product — which combines database and AI technologies — addresses some of the challenges encountered by data science teams while training, validating, and deploying AI systems. For example, it introduces native Spark structured streaming ingestion, a feature that makes streaming resources ostensibly easier to ingest than before. And Splice Machine 3.1 adds new database capabilities including foreign key processing, richer trigger support and improved handling, indexes on expressions, and improved import and export capabilities as well as DB2 compatibility.

Splice cofounder and CEO Monte Zweben says that the streaming ingestion capability should prove especially useful for industrial accounts connected to distributed control systems, where it’s essential to ingest data in real time as it becomes available. “With 3.1, we have made vital leaps in database capabilities,” Zweben said. “[They’ll] successfully operationalize real-time AI applications and bring machine learning models into production.”

Splice Machine 3.1 also aims to increase the transparency around the data used to create AI and machine learning models at scale. A new feature enables developers to query a database back in time with syntax to a specific date, providing an audit and lineage for a regulator checking for bias or data drift. It builds on Livewire, a product Splice launched last November that draws on live sensor data to predict operational problems to avoid outages and keep machinery up and running.

A number of enterprises in the process of adopting AI struggle with the key stages of data collection, preprocessing, and prep. A recent study found that less than 4% of companies report that the data used to train their AI systems presented no problems, with most data-related problems stemming from how the data was being produced and labeled internally. Biases or errors in the data and a lack of sufficient resources topped the list of data management problems with which businesses most often struggled.

“We are excited to be powering data engineers and data scientists with the tools they need,” Zweben added. “[We believe they’ll] break down the chasms that stop machine learning and AI projects from being successful.”

Splice Machine 3.1 is available as a fully-managed cloud service on Amazon Web Services, Azure, and Google Cloud Platform and is also available on-premises.