Everpure Strengthens Data Management Features for Large-Scale AI Operations

Everpure announced on the 1st that it is strengthening data governance, inference performance, and cost management capabilities. The aim is to support enterprise-scale AI operations. The update focuses on easing data accessibility, cost, and performance issues that arise as companies move AI from pilot stages to production environments.

The enhancement is based on Everpure's Data Primacy strategy, unveiled in June. Data Primacy is a concept that designs enterprise architecture around data and places data ahead of applications.

Everpure says it intends to provide a single management framework for the governance, security, and real-time access functions needed for AI agents to use distributed data. This aligns with an approach that consolidates management systems so AI can safely make use of distributed data.

MCP integration has been added to Everpure Data Intelligence. As a result, AI agents and security tools will be able to support natural language-based, real-time data catalog queries. Query items include checking the required data location and sensitivity classification.

Deployment has also been simplified. Related functions can be deployed from the existing Pure1 console without a separate management server. Along with this, the company is offering a privacy-first file intelligence feature. This feature allows users to verify access permissions and usage history without directly opening file contents.

AI inference performance improvements have also been added. The company provides Pure KVA(Key-Value Accelerator). The feature works by prepositioning FlashBlade storage context in GPU memory. This shortens TTFT.

Everpure focused on improving AI processing efficiency while responding to a multitenant environment in which data is not moved separately. Based on its own tests, Everpure said TTFT can be shortened by up to 20 times. It is also focusing on improving GPU utilization and increasing token throughput.

To improve storage efficiency, the company is applying DeepReduce. The method continuously analyzes storage blocks within FlashBlade and has the ability to identify data similarities that are difficult to detect with traditional deduplication methods. Through this, it can expand available capacity without separate work.

For AI cost management, Everpure also provides an open-weight model-based intelligent token optimization reference architecture. The architecture is designed to preserve a company's control over its own data, while aiming to reduce external AI service API token usage and improve predictability in cost management.

Everpure plans to respond to data access, security, AI inference performance, and cost management through a single operating framework. In this process, it will use sensitive data classification and access permission information for AI utilization, cyber resilience, and setting data recovery priorities. Related features will be rolled out sequentially starting in October.

Prakash Darji, Everpure's head of data and digital experience, said the reason enterprise AI falls short is not model performance but that data is not ready to support real-time autonomous agents, and added that Everpure will strengthen data access, governance, and automation so enterprise AI can expand beyond experiments into production environments.

Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407671

References

• https://www.epnc.co.kr/news/articleView.html?idxno=407671