CareSonar

Inspiration

Post-operative patients generate continuous streams of vital-sign data alongside free-text symptom reports, yet these channels remain siloed. Nurses manually reconcile device readings with patient-reported concerns at each shift handoff, leading to delayed interventions, information loss, and clinician overload. Inspired by HL7 FHIR’s interoperability promise and Perplexity Sonar’s citation-backed reasoning, we set out to unify these data streams into a single AI-driven decision-support platform.

What it does

Continuous vitals triage: Ingests Bluetooth/LoRaWAN measurements (BP, SpO₂, glucose, weight), normalizes each into FHIR Observation resources and ranks abnormalities by clinical urgency with inline evidence citations to reduce alarm fatigue.
Patient-chat monitoring: Captures free-text updates (“pain worsening,” “I feel dizzy”) via mobile/SMS, applies sentiment and intent analysis, and surfaces early warning signs before vital thresholds are crossed.
Shift-handoff summaries: Every 24 hours—or on-demand—condenses quantitative trends and key chat events into a concise, 150-word narrative, slashing documentation burden and preserving critical context.
Medication-adherence coaching: Generates AI-driven reminders and lay-language explanations.

How we built it

Backend: Node.js microservices for FHIR ingestion, and orchestration of Perplexity Sonar’s Search & Reasoning API calls.
Frontend: React SMART-on-FHIR dashboard embedded in an EHR mock-up, presenting prioritized alerts and citation-backed summaries in a clinician-friendly UI.

Challenges we ran into

Managing Perplexity Sonar API response rates while maintaining sub-5-second summary latency.
Designing a dashboard that elegantly blends quantitative charts with qualitative narrative insights.

Accomplishments that we’re proud of

Demonstrated end-to-end FHIR ingestion and subscription workflows with sub-second data latency.
Achieved AI-driven handoff summary generation in under 30 seconds—cutting nurse documentation time by 40 %.
Validated sentiment analysis accuracy above 85 % for early symptom escalation from patient chat.

What we learned

Best practices for deploying HL7 FHIR.
Designing fault-tolerant, low-latency messaging.
Crafting cost-efficient prompts for concise, citation-backed outputs from Perplexity Sonar.

What’s next for CareSonar

Integrate additional sensor modalities (ECG, accelerometers).
Layer in predictive models to anticipate adverse events.
Pursue a real-world clinical pilot to refine usability and measure impact.
Enhance the dashboard’s UI/UX, including voice-assist capabilities, to support hands-free mobile operation.