At Everpure’s Accelerate conference, we had the chance to catch up with General Manager for Digital Experience Prakash Darji. We talk about the additional capabilities the company announced around what it calls Universal Data Intelligence. Everpure wants to build an end-to-end AI data platform built on the principle that data, not applications, should be the foundation of enterprise architecture.
To be clear, Everpure is not abandoning its storage heritage. Darji is careful to frame Universal Data Intelligence as an addition, not a pivot. “This is not a shift in direction so much as an addition,” he explains. “We’re not running away from storage. We provide great storage. We’re continuing to make our storage better.”
What has changed is the recognition that AI has created an architectural paradigm shift. Applications no longer sit cleanly on top of isolated data stores. AI agents, large language models, and automation workflows require data from dozens or hundreds of systems, all at the same time. That data must be found, understood, governed, and prepared before it can be used effectively.
That’s where Universal Data Intelligence comes in. This is a framework that enables organizations to manage data and data policy independently of applications. In this framework applications become workflows running on top of a common data schema.
Data Intelligence and Data Stream
Universal Data Intelligence consists of two core components. The first is Data Intelligence. You can see this as the new name for 1touch, the solution by the company of the same name that Everpure acquired earlier this year. This handles discovery, cataloging, classification, and governance of data. The second is Data Stream, which handles vectorization and data preparation for AI consumption.
Data Intelligence
Not all data is equal. That is, some data is more reliable than other data. This is very important when AI agents start getting access to the data of an organization, Darji argues. He sees a diference between primary sources and secondary sources, which differ greatly in terms of reliability. As an example, he talks about an employee who downloads Salesforce data to a spreadsheet, merges it with finance figures, and tweaks it in a pivot table. An AI agent scanning that laptop file has no way of knowing it is not looking at a primary source.
Data Intelligence addresses this by cataloging all data across every source: block, file, and object storage, all major databases and file types, and SaaS platforms including Salesforce, ServiceNow, Workday, and Jira. It establishes what is a primary source versus a derived copy. It also applies attribute-based access controls and data governance. This should ensure that private or regulated information is handled appropriately before it reaches any AI model.
Using Data Intelligence, it is also possible to build ontologies to understand what data means in context. Using the example of a string like “ABC123,” Darji explains that without understanding the surrounding text, the relationship to other systems, and how the data is accessed and used, an AI model cannot determine whether that string is a product name, a random identifier, or something else entirely. Data Intelligence enriches data with attributes and builds a knowledge graph that tracks relationships across systems.
In other words, Data Intelligence creates a knowledge graph that is multi-dimensional, capturing degrees of relationship rather than simple categorical links. It is similar in concept to a recommendation engine that understands not just genre but the full web of connections between titles, actors, and viewing patterns.
Data Stream
Data Stream handles the next step in the AI data pipeline. It converts relevant data into vector-ready format for use with AI models. Vectorization introduces data bloat, which means that feeding all available data into a model is both expensive in terms of token costs and counterproductive in terms of accuracy. By integrating Data Intelligence with Data Stream, customers first identify which data is relevant and primary, then vectorize only that subset.
“Your results will be better by making sure relevant data goes into the model,” Darji says. “And the token costs will be lower because you’re not traversing a large blowout in the vectorization. So it’s more efficient from a token use and AI cost, and it’s more accurate.”
Using Data Stream in isolation, indexing everything without first applying Data Intelligence, is technically possible but not recommended. The integration between the two components is already built, and Everpure offers automated workflows to move customers through the pipeline, Darji says. These workflows do the following: identify relevant data, vectorize it, govern it, and deploy it on FlashBlade S for smaller environments or FlashBlade X for petabyte- and exabyte-scale deployments. In other words, Everpure wants to take care of the full stack.
A middleware philosophy in a world of walled gardens
Everpure’s approach is different from the strategies from many other vendors, like SAP, Salesforce, Databricks, and Snowflake, Darji says. These types of companies have built their respective businesses on effectively telling customers to move all their data into their respective ecosystems. Every enterprise already has most, if not all of these systems. This means moving a lot of data to a lot of different environments.
Darji not only notes that having many different environments with a lot of data isn’t efficient. He sees technical limitations too: “An analytics platform like Databricks or Snowflake holds transformed data optimized for querying, not transactional source data. If a business wants an AI agent to approve only profitable sales orders, the agent needs finance data from SAP and sales data from Salesforce and combine that. Neither system alone can provide both, and neither holds the source-of-record data the other needs.”
Data primacy frames an old idea in a new way
The framework and the resulting architecture Everpure advocates is a good example of thinking about enterprise architectures in a data-centric manner. The data doesn’t move, everything else is set up to accommodate what can be done with the data.
The concept of being “data-centric” has been circulating in enterprise technology for a decade, and Darji acknowledges that customers have heard it before. Everpure is deliberately reframing the concept as “data primacy”. This is to make it clear that data is the primary layer of enterprise architecture, with applications and AI agents operating on top of it rather than owning it.
Listen to or watch the full conversation to hear more details of what Everpure wants to achieve with Universal Data Intelligence, and how it wants to make sure it differentiates itself from other players in the market, who are also moving towards more headless architecture where the frontend doesn’t matter anymore.
Also read: From app-centric to open and data-centric: Can Everpure deliver on its promise?