We’ve talked about the critical role of centralized data platforms in enabling AI at scale for some time and in many issues of HDI. Those platforms—Databricks, Snowflake, Microsoft Fabric—remain the foundation for modern data strategy in healthcare. But recently, we’ve had the privilege of architecting new approaches with clients outside of healthcare that show us the next step forward.

These organizations gave us the flexibility to focus not just on building a solid core, but on driving ROI by extending that core with new capabilities. By layering Model Context Protocol (MCP) and Trino-based federated query architectures on top of their centralized data stacks, they unlocked outcomes that a single centralized platform could never deliver on its own.

The lesson is clear: healthcare systems can take the same step—pouring gasoline on their centralized platform investments, accelerating ROI, and expanding what’s possible.

The Plateau of the Centralized Data Platform

For many healthcare systems, the journey to a centralized platform has been long and complex. The payoff is real:

• A single source of truth through bronze, silver, and gold layers
• Consistent KPIs and dashboards for executives and clinicians
• Greater trust in data quality and governance

But there’s a growing frustration, too:

• Unstructured data is still sitting outside the platform. Clinical notes, scanned documents, images, and patient-reported outcomes are hard to ingest, classify, and activate.
• Federated queries across multiple operational systems remain painful. Teams want to combine financial, operational, and clinical signals quickly—but they’re forced into lengthy ETL projects.
• AI pilots stall. Without access to the right mix of structured and unstructured data, copilots and predictive models can’t move from proof of concept to production.

Healthcare systems aren’t questioning the value of their centralized platforms. But many are asking: Is this as far as we can go?

The answer is no.

What We Architected Outside Healthcare

Two recent projects—different industries, same challenge—demonstrated what’s possible. While we can’t share client names, the patterns are worth examining.

Architecture A: Unlocking Unstructured Content with MCP

We designed an MCP layer on top of an Azure SQL + Databricks backbone. This allowed the client to:

• Query unstructured and distributed content like contracts, PDFs, and scanned records without migrating them into the warehouse
• Surface insights directly into analytics and AI workflows, powered by a shared semantic layer
• Maintain governance by treating content as an accessible object with consistent policies, instead of a loose file system

The result? Their AI copilots could finally answer questions that relied on documents, contracts, and free-text—without compromising compliance or spending months on ingestion projects.

Architecture B: Federated Agility with Trino

For another client, we integrated a Trino-based query engine into their Databricks lakehouse. This gave them:

• The ability to join and query data across ERP, CRM, and IoT systems without replicating every dataset
• A federated query layer that respected centralized governance and role-based access
• Real-time, GenAI-enabled insights delivered directly in planning and operations tools

This wasn’t just about speed. It was about agility. Instead of waiting weeks for data engineering to build pipelines, business teams could access governed, near real-time insights on demand.

Both architectures kept the centralized gold-layer intact. But these extensions became accelerators—bridging gaps that the centralized stack alone couldn’t close.

Why Healthcare Needs This Next Step

Healthcare is sitting on the exact pain points these clients faced:

• Unstructured Data Explosion: Clinical notes, EHR attachments, insurance documentation, and patient surveys hold critical context for care and operations. Centralized stacks struggle to make this data usable at scale.
• Cross-Domain Questions: Finance, operations, and clinical questions rarely live in one dataset. Leaders need insights that span Epic, ERP, claims, and supply chain systems.
• AI Readiness: Every health system wants GenAI copilots for clinicians and operations staff. But copilots need curated, federated access to both structured and unstructured data to provide meaningful answers.

By layering MCP and Trino on top of Databricks, Snowflake, or Fabric, health systems can:

• Tap into new classes of data without massive migration projects
• Answer cross-domain questions in days, not months
• Empower AI copilots and command centers with governed access to a broader data universe
• Amplify ROI from the centralized stack they’ve already invested in

Architecture at a Glance

Here’s the blueprint we’ve used with clients outside healthcare, adapted for health systems:

1. Centralized Data Platform (Databricks, Snowflake, Fabric)

• Bronze → Silver → Gold medallion architecture
• Core governance, lineage, and security controls

2. MCP Layer

• Structured access to unstructured and distributed content
• Document-based governance, with audit and policy enforcement
• Enables GenAI copilots to pull answers from content repositories

3. Trino/Federated Query Layer

• Connects multiple operational systems (EHR, ERP, claims, IoT)
• Provides federated joins and near real-time analytics
• Ensures queries inherit centralized governance policies

4. Governance & Security

• Role-based and attribute-based access
• Zero-trust architecture with audit trails
• Unified policies across centralized and extended layers

5. Application Layer

• Clinical and operational GenAI copilots
• Real-time command centers
• Patient engagement portals
• Revenue and supply chain forecasting hubs

The ROI Impact

We architected these approaches not as side projects, but as accelerators for business value. The results were clear:

• Reduced Time-to-Insight: From weeks of pipeline building to near real-time query access
• Increased Trust: Governance extended seamlessly across new data domains
• Expanded AI Use Cases: Copilots and assistants that could finally work with the data users cared about most
• Higher ROI: Centralized platform investments amplified rather than duplicated

Healthcare systems could see the same outcomes—because the challenges, and the opportunities, are nearly identical.

Closing Thought

Your centralized platform isn’t the finish line. It’s the launchpad.

We’ve seen firsthand how layering MCP and Trino onto Databricks, Snowflake, or Fabric turns a strong architecture into a transformative one. For healthcare, this could mean unlocking clinical insights buried in notes, powering operational copilots that actually work, and accelerating ROI from investments already made.

The centralized gold layer remains the foundation. But with the right extensions, you can pour gasoline on that fire.