SafeGuard

Specialized clinical agent integrated into the PO ecosystem, leveraging SHARP Extension and A2A Protocol to provide seamless, evidence-based decision support directly within the clinician's workflow.

Comment

Inspiration

SafeGuard was born from a sobering reality: in high-pressure environments like hospitality and the tech industry, psychosocial risks remain "invisible" until they escalate into a crisis. We identified a massive disconnect between fragmented clinical data and executive decision-making. We were inspired to build a predictive bridge that transforms subtle behavioral signals into measurable clinical insights (ICD-10) and boardroom-ready economic value (ROI).

What it does

SafeGuard is an AI-powered Clinical Decision Support Agent fully integrated into the Prompt Opinion ecosystem. It acts as a specialized consultant for healthcare providers, offering:

  • Precision Risk Stratification: Categorizing patients into L0-L3 risk levels based on clinical scores (DASS-21, SRQ-20).
  • Intelligent ICD-10 Recommendations: Suggesting relevant Chapter V diagnostic codes with evidence-based clinical justifications.
  • Economic ROI Forecasting: Calculating real-time ROI estimates and INA-CBG claim potentials to justify the financial value of mental health interventions.
  • SHARP-Ready Analysis: Consuming FHIR context (Patient ID & Tokens) to provide context-aware analysis during active clinical sessions.

How we built it

  • AI Engine: Leveraged Google Gemini 2.0 Flash for high-velocity, clinical-grade reasoning and natural language processing.
  • Backend Architecture: Developed a robust Node.js & Express server implementing the A2A (Agent-to-Agent) v2.0 protocol.
  • Interoperability: Implemented SHARP Extension Specs to ensure standardized and secure handling of patient context.
  • Knowledge Base: Grounded our AI models in established clinical hierarchies and WHO-standard ROI modeling.
  • Deployment: Orchestrated on Render to ensure high availability and persistent streaming for complex medical data exchange.

Challenges we ran into

The primary hurdle was achieving perfect synchronization between our agent manifest (agent-card.json) and the Prompt Opinion platform parser. We underwent dozens of iterations to ensure our SHARP Extension and A2A Skills were correctly mapped within the UI. Additionally, transitioning from an SSE-heavy approach to a more stable REST-based A2A flow was critical to guarantee the integrity of sensitive medical data exchange in a serverless environment.

Accomplishments that we're proud of

We are incredibly proud to have built a "SHARP-Ready" agent that doesn't just "chat," but actually "communicates" with EHR (Electronic Health Record) ecosystems. Translating complex psychosocial distress into tangible ROI figures is a significant milestone, as it empowers healthcare leaders to move from reactive crisis management to proactive economic investment.

What we learned

This journey deepened our expertise in global interoperability standards like FHIR and SHARP. We learned that in healthcare AI, "intelligence" is nothing without "sovereignty." Designing an A2A flow taught us how to balance sharp predictive insights with the rigorous security needed to protect Protected Health Information (PHI) under regulations like HIPAA and Indonesia's PDP Law.

What's next for SafeGuard

Our roadmap includes:

  • Real-time EHR Write-back: Enabling direct ICD-10 suggestions into medical records via FHIR write operations.
  • IoT Ecosystem Integration: Connecting our proprietary "SafeGuard Companion" hardware to the A2A workflow for real-time biometric feedback.
  • Population Health Analytics: Developing long-term predictive models to track dynamic mental health trends across large corporate cohorts.

Built With

Share this project:

Updates