Docarion CareGraph & Rovo Clinical Triage Copilot

Atlassian Forge Hospital Incident Command, Rovo Clinical Triage Agent, Teamwork Graph & HIPAA-Safe MCP Server

Comment

Inspiration

In hospitals and multi-site health networks, clinical operations and IT/biomedical incident management live in dangerous silos. When a critical laboratory panic value fires in the ICU (such as potassium at 6.8 mmol/L or lactate at 5.4 mmol/L), when a neonatal ventilator enters battery failover, or when a blood-bank cold-chain sensor detects a thermal excursion, rapid response teams track the operational ticket in Jira Service Management (JSM) and Jira, while clinical Standard Operating Procedures (SOPs) and Corrective and Preventive Action (CAPA) postmortems live separately in Confluence.

On-call clinicians and biomedical engineers waste precious minutes switching between clinical systems, searching Confluence for the right hospital protocol, figuring out which specialist team is on call, and manually redacting Protected Health Information (PHI) before documenting root-cause postmortems. We built Docarion CareGraph & Rovo Clinical Triage Copilot on Atlassian Forge so hospital teams can build with clinical context and ship life-critical actions with confidence.

What it does

Docarion CareGraph & Rovo Clinical Triage Copilot is an Atlassian Forge application purpose-built for Healthcare (Apps for Specialized Industries) that unifies Jira, Jira Service Management (JSM), Confluence, Rovo Agents, Atlassian Teamwork Graph, and Atlassian Model Context Protocol (MCP):

  1. Jira & JSM Clinical Command Center (jira:issuePanel & jiraServiceManagement:queuePage):
    • Surfaces real-time NEWS2 Clinical Acuity Scores (0–20), SLA countdown timers, biomedical telemetry, and HIPAA/NDPR de-identification status directly inside hospital Jira issues and JSM incident queues (HOSP-401 ICU Sepsis & Hyperkalemia Panic Alert, HOSP-402 NICU Ventilator Failover, HOSP-403 Pharmacy Biologics Cold-Chain Excursion).
  2. Assignable Rovo Clinical Triage Copilot (rovo:agent + 4 Reusable action Modules):
    • action-triage-clinical-incident: Computes NEWS2 acuity and ESI level, identifies sepsis/electrolyte or biomedical hazard flags, and transitions the Jira/JSM incident priority and workflow status.
    • action-attach-confluence-sop: Queries the hospital's Confluence space via requestConfluence and Teamwork Graph to attach a structured, interactive Confluence Clinical SOP Checklist directly to the Jira issue.
    • action-generate-sbar-handoff: Synthesizes an evidence-bound SBAR (Situation, Background, Assessment, Recommendation) clinical handoff brief and dispatches escalation alerts to the on-call specialist team.
    • action-create-confluence-capa: Drafts and publishes a structured Joint Commission / ISO-15189 CAPA Postmortem Page in Confluence with automated 5-Whys root-cause analysis and preventive action items linked back to the Jira issue.
  3. Atlassian Teamwork Graph Healthcare Connector (graph:connector):
    • Ingests and links 5 healthcare entity types across the Atlassian Teamwork Graph: Jira/JSM Clinical Incidents, Confluence Clinical SOP Runbooks, Biomedical Equipment Assets, On-Call Specialist Teams, and Confluence CAPA Postmortems, giving Rovo and hospital teams 360-degree relational context.
  4. Atlassian MCP Server with Zero-Leakage HIPAA Safe Harbor & NDPR Guard (webtrigger + forge-app/mcp-server):
    • Exposes a JSON-RPC 2.0 Model Context Protocol (MCP) server with 5 healthcare tools (query_clinical_teamwork_graph, triage_hospital_jsm_incident, attach_confluence_clinical_sop, generate_sbar_handoff, create_confluence_capa_report) and deterministic PHI redaction that strips patient names, MRNs, phone numbers, emails, and national IDs before any payload is stored in Jira or Confluence.

How we built it

  • Atlassian Forge Platform (forge-app/manifest.yml & forge-app/src/index.ts): Built on Node.js 20.x runtime (nodejs20.x) using @forge/api (requestJira, requestConfluence, storage), @forge/resolver, and @forge/bridge.
  • Rovo Agent & Actions: Configured a specialized rovo:agent (docarion-rovo-clinical-copilot) backed by 4 deterministic action modules with readWrite and readOnly action verbs.
  • Teamwork Graph (graph:connector): Implemented docarion-clinical-graph-connector to map associative edges (GOVERNED_BY_SOP, IMPACTS_ASSET, ASSIGNED_TO_TEAM, DOCUMENTED_IN_CAPA) between Jira issues, Confluence pages, biomedical assets, and on-call teams.
  • Atlassian MCP Server (forge-app/mcp-server/index.ts & src/app/api/forge/mcp/route.ts): Implemented full JSON-RPC 2.0 MCP tools/list and tools/call handlers with an automated HIPAA Safe Harbor / NDPR PHI redaction pipeline.
  • Custom UI Healthcare Workbench (src/): Built with Next.js 14, React 18, TypeScript, Tailwind CSS, and Vitest unit tests verifying Teamwork Graph construction, deterministic Rovo action state transitions, and zero-leakage PHI scrubbing.

Challenges we ran into

  • Strict PHI De-Identification Across Multi-Product Context: Hospital incidents often contain free-text notes with embedded patient names, MRNs, or phone numbers. We engineered a deterministic HIPAA Safe Harbor & NDPR redaction layer that scrubs all 18 identifier patterns before any Rovo prompt, Jira comment, Confluence CAPA page, or MCP tool response is persisted.
  • Bridging Biomedical Telemetry with Clinical SOPs: Hospital incidents span both clinical lab emergencies and physical biomedical hardware failovers. Designing the Teamwork Graph schema to unify clinical protocols, asset calibration records, and on-call rosters allowed a single Rovo agent to reason seamlessly across both domains.

Accomplishments that we're proud of

  • Full-Stack Atlassian Ecosystem Unification: Combined Jira, Jira Service Management, Confluence, Rovo Agents, Teamwork Graph, and Atlassian MCP in a single cohesive healthcare solution.
  • Zero-Hallucination Evidence-Bound Clinical Triage: Every Rovo action is grounded in verified Confluence SOP runbooks and deterministic NEWS2 acuity scoring rather than unconstrained free-text generation.
  • Sub-Second SBAR Handoffs & One-Click Confluence CAPA Postmortems: Turned a 25-minute manual cross-system workflow into a 10-second guided flow inside Jira.

What we learned

  • Combining Teamwork Graph relationships with Rovo Actions dramatically improves agent reliability in regulated industries like healthcare because the agent traverses explicit links between an incident ticket, its governing Confluence SOP, and the responsible on-call team rather than guessing from keyword search alone.
  • Exposing Forge resolvers via Model Context Protocol (MCP) makes it effortless for external clinical AI assistants and hospital command centers to orchestrate Jira and Confluence workflows safely.

What's next for Docarion CareGraph & Rovo Clinical Triage Copilot

  • Expanding HL7 FHIR R4 and IEEE 11073 biomedical device event ingestion into the Atlassian Teamwork Graph connector.
  • Publishing Docarion CareGraph on the Atlassian Marketplace for hospital quality-assurance, clinical engineering, and rapid-response teams worldwide.

Built With

Share this project:

Updates

Submission history