Coder Legion - Recent posts tagged dataintelligence

The Purchase Order Is Canceled — and Executed. That's Not a Data Error.

Wed, 16 Sep 2026 12:30:00 +0000

An AI agent trying to answer "what did we invoice this customer" runs into a problem most demos never show: the honest answer isn't one number. There's what you invoiced in U.S. dollars, what that converts to in local currency, and the tax implication layered on top. There's a purchase order that got canceled and a separate one that got executed for the same order. To a system built to give a clean answer, that looks like corrupted data. To Naveen Neelakantam, product and technology strategy lead in Everpure's Office of the CTO, it's just what business data actually looks like — and building AI that tries to erase it is where a lot of agentic pilots start quietly failing.

"Contradiction is part of the data landscape," Neelakantam said in an interview about Everpure's approach to AI-ready data. "Those contradictions are meaningful. You might see a canceled purchase order plus the executed purchase order — we want to link all of those in the knowledge graph and say, if you're asking about POs, some of them are canceled, some are executed."

That's a deliberately different posture from a lot of "clean your data for AI" pitches, which treat conflicting records as noise to filter out before a model ever sees them. Everpure's approach, built on its Everpure Data Intelligence platform — acquired via 1touch.io, and already running in production for financial and insurance customers on-premises, not a fourth-quarter beta, per Neelakantam — keeps the mess and surfaces it instead.

The mechanics are worth understanding if you're the one building the agent. Data Intelligence scans structured sources like Snowflake, Salesforce, and SAP alongside unstructured ones — invoices, scanned purchase orders, object stores — looking for entities: names, addresses, PO numbers. "In these invoices or this OCR document, I think this is a name, I think this is a PO number, I think this is an address," Neelakantam said. It then checks those candidate entities against a canonical source a business has designated as reliable, such as a Salesforce record, and links matches into a knowledge graph. Where two systems disagree, both versions stay in the graph, tagged as related, rather than one getting silently discarded.

That matters more as agents start reasoning across systems instead of staying inside one application's walls. Neelakantam framed the shift as less about tearing down applications than about what sits above them: an agentic pricing engine, for example, isn't really an application in the traditional sense. It pulls from several systems, gets trained on success criteria, and plugs back in through APIs. For that to work reliably, the data underneath has to carry more context than a database column ever did on its own. Vocabulary is part of that problem. "All words are not created equal," he said. The word "inventory" means something different to a finance team than to a shipping team, and grounding that word in the actual documents and systems where it shows up, rather than treating it as plain English, is what lets a model tell the difference.

None of this means foundational models get it right by default once the data's connected. Neelakantam pointed to case-deflection systems as a concrete failure mode: an agent auto-closing a support ticket as resolved when the customer is still angry. "Nobody likes having a case auto-closed on them," he said. Fixing that kind of error isn't a matter of a new rule or a line of code, in his view. It takes the same discipline as traditional machine learning: logging inputs and outputs, running feedback loops, tracking precision and recall, and in some cases fine-tuning on the right training data. Prompt tuning alone doesn't get a system to production-grade accuracy.

The audit trail follows the same logic. As agents take on more autonomous action, Neelakantam said the industry is only beginning to build out what detailed logging of agentic workflows actually looks like — capturing intent, reasoning, and outcome at every step, not just the final answer. For regulated industries already running Data Intelligence in production, that's not optional. It's the same instinct driving the rest of this approach: don't hide the parts of the system that are hard to explain. Log them.

Neelakantam was candid that this transition carries real risk. Coding has gotten more accessible faster than the engineering discipline needed to deploy AI systems safely has caught up, in his view — a "wild west" period where people can build and ship agentic systems without necessarily understanding the failure modes underneath them. His read isn't that this slows down. It's that developers need to get comfortable fast with logging, testing, and documentation as the basic hygiene of handing real responsibility to a system that reasons on its own.

For developers building on top of any AI-readiness platform, not just Everpure's, the takeaway is less about the specific product and more about the posture: treat contradictory data as a signal, not a bug to be squashed before a model ever sees it. The version of your data with all its disagreements intact is more honest than the version an engineer smoothed over, and it's the version that will keep an agent from being confidently wrong.

From Storage Vendor to Data Partner: A Conversation with Everpure's Nirav Sheth

Thu, 18 Jun 2026 21:30:43 +0000

The first thing a company typically discovers when it deploys Everpure Data Intelligence isn't what it expected. It's data it didn't know it had.

Nirav Sheth, VP of Worldwide Systems Engineering at Everpure, described a proof of concept that illustrates the pattern well. A customer connected Data Intelligence to their environment and found sensitive financial data sitting in a Google Drive folder with open permissions — visible to anyone in the company, with no governance controls of any kind. Nobody had flagged it. Nobody had even known it was there.

"I ask customers: do you have a really good handle of all your data across all your applications from the edge to the data center to the cloud?" Sheth told me at Pure Accelerate 2026. "And many actually just start to laugh as an acknowledgment of, yeah, you got me."

That moment — when a customer realizes the scope of what they don't know — is where Sheth's work actually begins. As VP of Worldwide Systems Engineering, he's closer to real-world deployment than the product managers and executives who set roadmap direction. What he sees most often isn't a technology problem. It's an awareness problem.

What the First 90 Days Actually Look Like

When customers first deploy Data Intelligence, the expected conversation is about AI readiness, pipeline automation, and semantic context. What actually happens is simpler and more unsettling: the tool maps the full scope of the customer's data estate, and that scope is almost always larger and messier than anyone anticipated.

Sheth described the sequence clearly. Step one is discovery — understanding what data exists, where it lives, and who can access it. Step two is classification and contextualization — tagging sensitive data, building governance policies, and mapping relationships across systems. Only after those two steps does the conversation about AI pipelines and Data Stream actually make sense, because only then does the customer understand what should and shouldn't be feeding those pipelines.

"Some customers find they actually have to do a lot of data cleanup before they even think about a data pipeline," Sheth said. "You need to make sure the right things are actually going to be going into that pipeline."

The court system demo staged during Wednesday's keynote illustrated the point. A large metropolitan court system — Sheth declined to name it — used Data Intelligence to map its entire data estate across multiple tiers of courts before building any AI models. The reason: too much sensitive data to use any third-party AI service. They needed to understand what they had, wall off what couldn't be touched, and build a secure internal capability — the equivalent of their own Harvey AI — before any pipeline work could begin. Two months after not knowing what data they had, they were extracting value from it.

The Global Manufacturer: What Actually Happened

One of the more detailed customer stories to come out of this week's conference involves a global manufacturer with 80 sites worldwide. Sheth walked through the full arc of that engagement during our conversation, and it's worth being specific about what the case study actually shows — and what it doesn't yet.

The manufacturer was running legacy hyperconverged infrastructure across all 80 sites, with support approaching its end. The security implications were immediate: once the vendor stops issuing updates, every unpatched vulnerability becomes a permanent exposure. That alone was a compelling reason to modernize. But the performance case was equally clear — hyperconverged architecture was struggling to keep pace with production throughput as manufacturing volumes increased.

The engagement ran 60 days, start to finish. The first phase covered the self-assessment, discovery, and transition planning — mapping current-state architecture and modeling what a move to a three-tier architecture based on Everpure FlashArray would look like at scale. The second phase went further: exploring what a video surveillance solution could do for manufacturing quality. The answer, it turned out, was significant. By building AI models around video feeds on the manufacturing line, the company could detect counterfeit parts in real time, reducing defect rates and improving field quality.

One thing worth clarifying: the gross margin expansion Sheth referenced in his keynote presentation is a projection, not a delivered result. The migration plan for 80 global sites is approximately 1 year for phase 1. The business case has been made to the CISO and CIO — and it was strong enough for them to commit — but the results are still being built.

What struck me about the story was Sheth's description of how the engagement was framed. "We anchored on what's most important for them," he said. "Production is money. We really wanted to anchor around what's going to ensure that they have higher production volume, better throughput, and fewer quality issues — versus coming in and saying, hey, we have a really cool technology platform."

That framing — infrastructure partner as business consultant rather than hardware vendor — came up repeatedly throughout our conversation.

The Migration Risk Nobody Talks About

The most honest answer Sheth gave came when I asked about the most common architectural mistake systems engineering teams see when customers try to scale before addressing existing risk.

His answer wasn't what I expected. He didn't describe a technical anti-pattern or a common configuration error. He described organizational inertia. "One of the biggest general challenges is sometimes a customer has been with a certain platform for five years, and they feel like moving to a different provider could end up being a bigger challenge or a bigger risk than staying with the incumbent."

His point is that the decision not to move is itself a risk posture — one that customers often don't examine critically because it feels like the safe default. Everpure's response to that has been to build SLA-backed migration tooling and use Data Intelligence to map what's running on the incumbent environment before any transition begins, making the migration plan more defensible and reducing the perceived risk of change. The goal is to earn trust incrementally, demonstrate value on the customer's existing infrastructure, and let the case for deeper adoption build from there.

"We want to meet them where they're at, we want to earn their trust, and then over time they'll continue to grow with us," Sheth said. The 84 NPS score — independently audited — is the metric he cited as evidence that the approach is working.

The Blueprint as Sequencing Tool

Sheth's framing of the EDC Success Blueprint differs slightly from how it was presented in formal announcements, and the difference is worth surfacing for engineering leaders evaluating whether to engage with it.

Officially, the Blueprint is a structured methodology — self-assessment, maturity guides, dedicated workshop — for moving toward an Enterprise Data Cloud architecture. In practice, Sheth describes it less as a methodology and more as a sequencing tool. Its real value is in helping organizations with 15 competing priorities figure out which ones to tackle first, in an order that's actually achievable given their current state, available resources, and internal political realities.

"If you try to take on too much in any one go, you might not have the level of internal support that you need," he said. "We really want to make sure that we're anchoring on what's most important in terms of our customers' priorities and their business outcomes."

The workshops included with Evergreen subscriptions are on-site, dedicated sessions — not a fee-based add-on. Sheth was clear about that distinction because it affects how customers should think about the commitment being asked of them. Everpure is investing its own systems engineering resources in these engagements, which means the company has a stake in the outcome being real rather than just the contract being signed.

A Different Company Than It Was

Sheth's closing comment was the most direct summary of what this week has actually been about, beyond any individual product announcement: "Everpure is a very different company than where we were a couple of years ago."

He pointed specifically to the company's founder, echoing the same sentiment on stage: the company that built its reputation on FlashArray has expanded into a full-stack data intelligence platform where storage is the foundation, not the ceiling. Data Intelligence, Data Stream, Portworx — these products add value regardless of whether the underlying storage is Everpure's. That's an intentional architectural bet, not a side effect.

"We can now add value no matter where the data resides," Sheth said.

For developers, engineers, and architects deciding whether to pay attention to Everpure's announcements this week, that's the most useful frame. The company is no longer selling storage performance as its primary value proposition. It's selling visibility and control over data that enterprises already have, wherever that data happens to live — with Everpure storage as the preferred home but not the prerequisite for value.

Whether that bet pays off at scale, across 14,500 customers with varying levels of maturity and readiness, is the story that plays out over the next several years.

Everpure (NYSE: P) provided press access to Pure Accelerate 2026.

Inside Everpure's Data Intelligence: A Conversation with Prakash Darji

Wed, 17 Jun 2026 22:14:14 +0000

When ransomware hits, the standard playbook is to roll back the entire environment to the last known clean snapshot. It works, but it's blunt. You lose every legitimate change made since that snapshot, not just the malicious ones.

Prakash Darji, General Manager of Digital Experience at Everpure, described a more targeted approach during a conversation at Pure Accelerate 2026. Instead of rolling back everything, Data Intelligence can identify exactly which files were affected by an attack and execute a micro-rollback — restoring only the compromised data while leaving everything else untouched.

It's a small example with a big implication: once you actually understand what your data is and how it connects to everything else, you can act on it with precision instead of brute force.

That's the premise behind Everpure Data Intelligence, the product formerly known as 1touch.io, and the centerpiece of what Everpure is calling data primacy — a fundamental shift in how enterprise data architecture should work in the AI era.

How the Semantic Layer Actually Works

It's easy to talk about "semantic knowledge graphs" in the abstract. Darji walked through what it actually looks like in practice, using a purchase order as an example.

A PO might start as a file sitting in a file share. That file lives inside a database, which follows a particular schema, which contains specific fields. A content classifier identifies what those fields actually represent — this field is a vendor name, that one is a dollar amount. All of that gets mapped into a knowledge graph that captures the relationships: this purchase order connects to a specific entry in Oracle, which connects to a specific transaction in the billing system.

That's seven layers of context — file share, database, schema, field, content classifier, knowledge graph, relationships — built around a single piece of data that, in a traditional application-centric architecture, would just be an opaque record sitting in a silo.

The payoff is that once that semantic structure exists, AI agents can actually reason across systems instead of guessing. An agent looking at a purchase order can understand not just what the document says, but how it fits into the broader business process around it — without a developer manually writing integration logic to connect Oracle to the billing system every time someone needs that context.

Two Systems, Converging

One detail worth flagging for anyone tracking Everpure's roadmap: the fleet telemetry data Everpure has collected from its storage systems for years, and the semantic layer that Data Intelligence provides, are currently separate systems. According to Darji, they're expected to converge within the next year.

That's not a small integration. Everpure's storage telemetry operates at the infrastructure level — looking at volumes, file systems, and performance data. Data Intelligence operates at a higher level — understanding what the data means and how it relates across applications. Bringing those together means an AI agent could eventually reason about both what a piece of data represents and how the infrastructure underneath it is actually performing, in a single context.

It's a useful reminder that data primacy isn't a single product launch. It's an integration project that will play out over the next several quarters, built on a still-recent acquisition.

Small Models, Better Data

One of the more counterintuitive points Darji made — echoed elsewhere at the conference — is that small language models with focused, relevant data are outperforming large language models fed raw, fragmented data. That cuts against the assumption that bigger models always win.

The logic follows directly from the semantic layer argument: if an AI system has to wade through unstructured, poorly contextualized data to find what's relevant, it needs a much bigger model and a much bigger context window to compensate. If the data is already classified, contextualized, and connected through a knowledge graph, a smaller, more efficient model can get better results with a fraction of the token cost.

For engineering teams evaluating AI infrastructure investment, that's a meaningful trade-off to understand. The instinct to solve data fragmentation by throwing a bigger model at the problem may be solving the wrong layer entirely.

The Adoption Gap

Darji was candid about where most customers actually stand. By his estimate, 80 to 90% of customers still don't fully understand semantic data management — what it means, why it matters, or how it changes the way they should think about their data architecture.

That's a notable admission from a product leader at a company betting heavily on this direction. It also explains why so much of the messaging at this conference, from the keynote through every press conference session, spent as much time explaining the concept of data primacy as it did describing the products that implement it. Everpure is still in the education phase of this transition, not just the deployment phase.

Why Bet Horizontal

Darji was direct about Everpure's competitive logic: Data Intelligence works across any storage, not just Everpure's own platform. That's a deliberate choice, not an oversight. Everpure isn't using Data Intelligence as a wedge to sell more hardware — it's positioning the product as a horizontal semantic governance layer that adds value whether or not a customer runs Everpure storage underneath it.

He compared it directly to the company's earlier bet on all-flash storage. There were plenty of competitors in that race too, and Everpure backed the bet anyway because it believed the architecture was right, not because the path was uncontested. The same logic applies here: data primacy is a horizontal architectural bet, and Everpure is betting that being right about the architecture matters more than controlling every layer of the stack.

From Application GUIs to Persona-Based Workflows

The most forward-looking part of the conversation was Darji's prediction about how enterprise software itself changes once a shared semantic layer exists.

Today, most enterprise software is organized around applications — you log into your CRM to do CRM things, your ERP to do ERP things, and switching between them means manually carrying context in your own head. Darji's prediction is that this shifts toward persona-based workflows: instead of organizing around which application owns which task, software organizes around what a specific role actually needs to accomplish, pulling from a shared semantic schema regardless of which system the underlying data lives in.

In that world, the work shifts for engineers too. Less time goes into building one-off integrations between specific applications, and more time goes into managing schema consistency — making sure the shared semantic layer accurately represents what the business actually means by its data, so that every persona-based workflow built on top of it stays trustworthy.

It's a meaningful shift in what application development looks like, and it's a direct extension of the data primacy argument Everpure has been making all week: once data carries its own context, the application stops being the thing organizing the work.

The Pragmatic Reality of Migration

None of this means existing enterprise systems disappear overnight. Darji's view here is pragmatic, echoing a line he attributed to a former PeopleSoft CEO: nobody builds the world in seven days when there's already an install base to deal with.

Nobody is ripping out SAP. The realistic path is that new workloads get built data-first from the start, while legacy complexity gradually thins out over time as the economics and the use cases shift. Migration is a multi-year process, not a single cutover — a point Charlie Giancarlo also made directly when discussing Everpure's own roughly two-and-a-half-year internal journey toward data primacy.

The Practical Takeaway

For developers, engineers, and architects evaluating where to invest attention, Darji's conversation reinforces a theme that's run through this entire conference: the bottleneck for enterprise AI isn't model capability. It's whether your organization actually understands its own data well enough to feed AI systems something trustworthy.

Data Intelligence is Everpure's bet that solving that problem requires a dedicated semantic layer — not a bigger model, not another data lake, and not another copy of the same fragmented data. Whether that bet pays off depends on execution over the next several quarters, as Everpure works through its own admission that most of the market isn't there yet either.

Everpure (NYSE: P) provided press access to Pure Accelerate 2026.