Enhancing Healthcare Service Delivery with RAG and LLM-Powered Search

accolade · Healthcare · year n/a · source Agentic: yes · promotional bias: high · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Healthcare service data access

Accolade needed to help internal healthcare teams answer complex member inquiries quickly despite fragmented data across multiple systems and strict governance requirements.

Constraints:

Healthcare data was fragmented across platforms such as AWS Redshift and Snowflake.
Siloed data hindered real-time member stratification and timely care delivery.
Healthcare compliance required governance, access controls, and lineage.
Internal teams needed access to PDFs, protocols, contracts, and customer information.
AI initiatives were delayed by the complexity of managing multiple data systems and models.

Architecture

Accolade consolidated data into a Databricks lakehouse, governed it through Unity Catalog, and built a RAG solution with Agent Bricks Custom Agents and DBRX. Internal users can query diverse data sources such as PDFs, online protocols, and customer information; DBRX generates context-aware outputs, and Databricks Model Serving exposes the model through an API with versioning, autoscaling, and lifecycle management.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  healthcare-data-sources["Healthcare data sources"] --> databricks-lakehouse["Databricks lakehouse"]
  databricks-lakehouse --> unity-catalog["Unity Catalog"]
  databricks-lakehouse --> agent-bricks-rag["Agent Bricks RAG"]
  unity-catalog --> agent-bricks-rag
  agent-bricks-rag --> dbrx["DBRX"]
  dbrx --> model-serving["Model Serving"]
  model-serving --> internal-decision-systems["Internal decision systems"]
  internal-decision-systems --> healthcare-data-sources

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Healthcare data sources`	Fragmented structured and unstructured data from platforms, documents, protocols, and customer information.	`Databricks lakehouse`
`Databricks lakehouse`	Unified storage and analytics environment for Accolade data.	`Unity Catalog`, `Agent Bricks RAG`
`Unity Catalog`	Governance, access controls, data lineage, and HIPAA-supporting controls.	`Agent Bricks RAG`
`Agent Bricks RAG`	Retrieval-augmented workflow over proprietary enterprise documents and data.	`DBRX`
`DBRX`	Open-source LLM used to generate accurate and context-aware outputs from proprietary data.	`Model Serving`
`Model Serving`	Deploys the model as a RESTful API with version control, autoscaling, and lifecycle management.	`Internal decision systems`
`Internal decision systems`	Care-delivery and service workflows that use model outputs to answer inquiries.	`Healthcare data sources`

Data flow

Structured and unstructured healthcare data flows into a governed Databricks lakehouse. Unity Catalog manages access and lineage, Agent Bricks builds a RAG workflow over PDFs, protocols, contracts, and customer information, DBRX generates answers, and Model Serving exposes the model to decision systems.

“Central to this implementation was a lakehouse architecture that enabled the combination of data storage and management to facilitate easier access and analysis.”article, stated

“to develop a RAG solution specifically tailored to improve the efficiency and effectiveness of their internal teams.”article, stated

Why this architecture

Accolade first unified and governed its data because RAG over healthcare information depends on trusted, accessible, compliant source context. DBRX and Model Serving then provided a controlled path to generate internal answers and integrate them into decision systems.

“This unified environment allowed Accolade to break down the silos between different types of data and systems”article, stated

“was critical for management and governance, supporting HIPAA compliance requirements with stringent access controls and detailed data lineage.”article, stated

Models used: DBRX · Databricks Agent Bricks Custom Agents · Databricks Model Serving · Apache Spark · Unity Catalog

Scale

Accolade data was spread across multiple platforms including AWS Redshift and Snowflake.
“the complexity of managing and coordinating multiple models across various systems, such as AWS Redshift and Snowflake”article, stated
Databricks enabled work at a scale Accolade previously could not manage.
“Databricks enabled us to do effectively at a scale we previously couldn’t manage.”article, stated
The platform governs structured and unstructured data, ML models, notebooks, dashboards, and files across any cloud or platform.
“govern their structured and unstructured data, machine learning models, notebooks, dashboards, and files across any cloud or platform.”article, stated

Alternatives considered (roads not taken)

Continue operating fragmented data systems.

Rejected because: Silos slowed response times and reduced the ability to personalize timely care.

“Fragmented data across multiple different platforms hampered our efficiency and ability to provide timely, personalized care.”article, stated

Build AI on top of the existing complex BI/data landscape.

Rejected because: The fragmented landscape delayed AI initiatives, so Accolade moved to an AI-driven framework.

“the company’s complex data management landscape also complicated the development of AI-driven initiatives”article, stated

Patterns

`rag-retrieval` (retrieval)

Accolade uses RAG over PDFs, protocols, customer information, and enterprise data to answer internal questions.

“retrieval augmented generation (RAG), a generative AI workflow that uses custom data and documents to provide context for LLMs.”article, stated

`privacy-preserving-processing` (governance)

Unity Catalog governs healthcare data access, lineage, and HIPAA compliance.

“supporting HIPAA compliance requirements with stringent access controls and detailed data lineage.”article, stated

`tool-use-orchestration` (deployment)

Databricks Model Serving deploys the LLM as an API with lifecycle controls.

“It deployed the model as a RESTful API, enabling real-time predictions that could be integrated directly into Accolade’s decision systems.”article, stated

`observability-tracing` (governance)

The platform provides transparency, control over costs and resources, and lifecycle management for ML models.

“greater transparency and control over resources and costs to simplify the development and maintenance of ML models”article, stated

Tradeoffs

Data foundation

Chose Unified lakehouse architecture. over Continuing fragmented BI and data platforms..

Rationale: Unified data made AI use cases actionable and efficient.

“It’s not just about having data; it’s about making it actionable and efficient”article, stated

Model serving path

Chose RESTful API through Databricks Model Serving. over Ad hoc model integration into decision systems..

Rationale: Model Serving provided version control, autoscaling, integration, and lifecycle management.

“With Model Serving, the healthcare technology company had Databricks’ support for version control, automatic scaling and integration for comprehensive lifecycle management.”article, stated

Gotchas (what bit them)

Fragmented healthcare data slowed care support.

Root cause: Internal teams could not quickly access necessary information across multiple platforms.

Resolution: Unify data in a Databricks lakehouse and use RAG for internal access.

“Our internal teams struggled to quickly access the necessary information, leading to slower response times”article, stated

Healthcare RAG requires governance before generation.

Root cause: Sensitive structured and unstructured healthcare data must be governed with access controls and lineage.

Resolution: Use Unity Catalog for governance and HIPAA-supporting controls.

“supporting HIPAA compliance requirements with stringent access controls and detailed data lineage.”article, stated

RAG quality depended on proprietary enterprise context.

Root cause: Internal users needed accurate answers from contracts, protocols, customer information, and documents.

Resolution: Use DBRX with proprietary enterprise data for context-aware outputs.

“DBRX to enhance internal search functions and enable internal users to retrieve accurate information quickly.”article, stated

Evaluation

The source does not publish benchmark scores; it reports operational evaluation through faster information access, productivity gains, governance, cost/resource transparency, and improved ability to answer complex member inquiries.

Metrics: Productivity gains · Response-time improvement · Governance and compliance readiness · Resource and cost transparency

“With the help of generative AI, the Accolade team developed solutions to quickly answer complex member inquiries, bolstering major productivity gains.”article, stated

Outcomes

Unified fragmented healthcare data into a governed lakehouse.
Enabled internal teams to answer complex member inquiries more confidently.
Improved access to PDFs, protocols, customer contracts, and customer information.
Created a governed foundation for further healthcare AI initiatives.

Accolade moved multiple systems including AWS Redshift and Snowflake into a governed lakehouse that spans structured/unstructured data, models, notebooks, dashboards, and files.

“The integration of RAG tools continues to support Accolade’s internal teams, providing them with the confidence to handle complex queries”article, stated