industry-department-solutions
latest
false
- Overview
- Medical Record Summarization
Healthcare Solutions user guide
Last updated Apr 20, 2026
System components
Medical Record Summarization (MRS) is built on several UiPath platform services that work together to ingest medical records, generate clinical summaries, and store results for downstream access. This page describes each component, the Data Fabric entities MRS provisions in your tenant, the storage buckets it uses, and the OAuth scopes required by the application.
UiPath platform components
| Component | Type | Role |
|---|---|---|
| MRS Pro-code Application | React 19 SPA (Vite + TypeScript) | User interface for file ingestion, clinical review, template authoring, test runs, export, and administrative configuration |
| MRS Run Job Process | Orchestrator process | Job scheduler that distributes work across N concurrent summarization workers using modulo-based assignment |
| MRS Agentic Process | Orchestrator agentic BPMN process | Per-worker orchestrator: queries PdfPackets, retrieves files, invokes the MRS Agent, writes results to Data Fabric, and deletes processed files |
| MRS Agent | AI agent | Core summarization agent: performs JIT DeepRAG context grounding against the source PDF and calls the Large Language Model (LLM) to generate a structured clinical summary with inline citations |
| MRS Discovery Storage Bucket | Orchestrator process | Scans the ingestion storage bucket on a cron schedule for files newer than the last processed date, creates PdfPackets records, and starts the Run Job process |
| MRS Agent - Test Prompts | AI agent | Runs template prompts against persistently indexed sample files to validate output quality before a template version is published |
| MRS Agent - Generate Prompts | AI agent | Reverse-engineers system and user prompts from an uploaded example summary, used to seed a new template |
| Agentic - Update Index | Orchestrator agentic process | Ingests sample files into Enterprise Content Services for persistent DeepRAG indexing (2-hour timeout, 30-second polling interval) |
| Agentic - Test Prompts | Orchestrator agentic process | Orchestrates test prompt runs: reads template version prompts, invokes MRS Agent - Test Prompts against the persistent index, and saves results to MRSTemplateTestResultOutput |
| Agentic - Generate Prompts From Summary | Orchestrator agentic process | Downloads the uploaded example summary and invokes MRS Agent - Generate Prompts to reverse-engineer system and user prompts, then saves results to MRSTemplateInitialPrompts |
| Agentic - Get Index Status | Orchestrator agentic process | Polls Enterprise Content Services to check persistent index ingestion status during the Update Index flow |
| Context Grounding (DeepRAG) | UiPath AI service | Grounds the LLM against medical record content. Used in JIT mode for production summarization and Persistent Index mode for template testing |
| Enterprise Content Services | UiPath AI service | Indexes sample files for persistent DeepRAG queries during template testing |
| GenAI Activities | UiPath AI service | Generates clinical summaries and citations from grounded context |
Data Fabric entities
MRS provisions 12 entities in your Data Fabric tenant at deployment time.
| Entity | Purpose |
|---|---|
PdfPackets | One record per source document queued for processing. Tracks status, modulo assignment, template ID, and ingestion source |
MRS_Output | One record per completed summarization. Contains generated summary files, citations, override content, reviewer assignment, OutputStatus, and timestamps. See MRS output data |
MRSTemplate | Master template records |
MRSTemplateVersion | Versioned template configuration: system prompt, user prompt, concurrent job count, and published status |
MRSApplicationConfig | Application-level configuration including feature flags and developer overrides |
MRSManualBatch | Tracks manual batch uploads initiated through the application |
MRSIngestionSource | Defines the storage bucket name and root path for automated discovery deployments |
IngestionState | Tracks the last processed folder date for scheduled discovery, enabling incremental processing |
MRSPacketSamples | Sample files uploaded for template testing, including indexing status |
MRSTemplateInitialPrompts | AI-generated prompts produced by the Generate Prompts agent from an uploaded example summary |
MRSTemplateTestResultOutput | Results from test prompt runs, stored per template version |
MRSAssignmentHistory | Audit record of reviewer assignments and reassignments for each MRS_Output record |
Storage buckets
Three Orchestrator storage buckets are provisioned within the Orchestrator folder selected for MRS during deployment.
| Bucket | Contents | Notes |
|---|---|---|
MRSMainBucket | Ingested medical record PDFs awaiting or in-progress summarization | Files are deleted by the cleanup workflow after summarization completes. For automated discovery, organize files in date-based subfolders (YYYY/MM/DD/) |
MRSSamplesBucket | Sample PDFs uploaded by template authors for persistent index testing | Managed through the template editor |
MRSSummarizationPackets | Output packet archives produced by the summarization pipeline | Read by the export functionality in the MRS application |
OAuth scopes
The MRS application (MRSApp) is registered as a non-confidential OAuth application in your UiPath tenant and uses PKCE for authentication. The following scopes are required.
| Scope | Service |
|---|---|
Insights | Insights |
PM.Group PM.Group.Read PM.User PM.User.Read | Platform Management |
OR.Administration OR.Execution OR.Folders OR.Jobs OR.Machines OR.Robots OR.Settings | Orchestrator |
DataFabric.Data.Read DataFabric.Data.Write DataFabric.Schema.Read | Data Fabric |