Acceldata can now spot data anomalies across multiple dimensions

Data observability vendor Acceldata Inc. today announced Adaptive AI Anomaly Detection capabilities in its xLake Reasoning Engine that automatically identify hidden, multidimensional data anomalies before they can disrupt business operations.

Launched in February, the xLake Reasoning Engine is a core component of the company’s Agentic Data Management platform. It operates across various environments, including hyperscale clouds, data clouds and on-premises systems, functioning as an artificial intelligence-aware data processing engine that integrates governance and security features.

The company said traditional anomaly detection tools identify one-dimensional errors such as a misplaced zero in a sales figure. Adaptive anomaly detection can also spot hidden anomalies across multiple data dimensions, such as a credit card account that lists transactions in locations hundreds of miles apart at the same time.

Chief Executive Rohit Choudhary (pictured) said such capabilities are particularly important now that large language models are being increasingly trained with structured data to ensure accuracy and validation.

“That means rollbacks are hard, and it’s difficult to change the nature of inferencing once you’ve fed in the data,” he said.

Narrowing the surface area

Data volumes are growing so quickly that conventional quality and validation checks can no longer keep up, he said. “That means the surface area has to be narrowed to issues that show anomalies,” he said. “There are too many correlating factors that affect behavior. You can’t wait for something to tell you what happened after the fact, so operational effectiveness is determined by finding hotspots.”

XLake can simultaneously evaluate anomalies across multiple attributes such as sales, product ID, region and time with the ability for customers to prioritize high-risk data segments for better performance. It detects unique patterns that static rule-based systems cannot and adapts continuously without manual tuning, Acceldata said.

“We identify high-risk data segments automatically and can identify anomalies in sensitive states or websites you provide,” Choudhary said.

In areas such as fraud detection, “systems will get extremely advanced because any piece of information has a context,” Choudhary said. “The more interactions you can track to find out something is wrong, the faster you can act.”

Acceldata collects metadata and monitors signals from data, pipelines, infrastructure, users and costs. Multivariate anomaly detection uncovers interdependencies that the company said traditional tools often miss. Agents can be configured to take actions such as forecasting business impacts and issuing compliance alerts, with the option of automated remediation.

Root-cause correlation links infrastructure failures with pipeline breakdowns and data spikes to pinpoint root causes. System can also correlate budget overruns to specific workloads, users, queries or processes; detect unusual access patterns by correlating user identity, location and data sensitivity; connect upstream data issues with downstream analytics; and identify early signals of delays by correlating processing times, data volumes and resource constraints.

Remediation capabilities were designed with human-in-the-loop in mind. “We make sure humans are present all the way from configuration to outcome management,” Choudhary said.  “Agents come with set of capabilities and prompts out-of-the-box but we give you ability to remove or modify them.”

The agentic data management platform will launch next month.

Photo: SiliconANGLE