CareLink+

• 

Every year, thousands of patients are discharged from hospitals only to face preventable post-operative complications at home. Most of these issues start with subtle symptoms — a missed dose, a mild fever, or an unusual heart rate — that go unnoticed until it’s too late. We wanted to change that.

CareLink+ was born from the idea that voice could be the bridge between patients and proactive care. By turning routine daily check-ins into actionable data, we empower caregivers to intervene early — before small issues become emergencies.

What It Does

CareLink+ is a voice-first, AI-powered post-operative care management system that transforms patient check-ins into real-time health insights.

Key capabilities include:

🎙️ Voice Interaction (Vapi + Fish Audio): Patients can talk naturally in English or Spanish, logging how they feel and reporting symptoms.

⚕️ AI Triage Engine (Claude): Analyzes responses, vitals, and medication adherence using evidence-based thresholds to assign risk levels (0–3).

📊 Dynamic Dashboard: Displays vitals, trends, notes, and risk factors using real-time analytics (Elastic + Kibana).

🔔 Automation Workflows (Composio): Sends alerts to caregivers, schedules medication reminders, and even auto-generates PDF reports.

🎥 LiveKit Video Conferencing: Instantly opens virtual “care huddles” when high-risk events are detected.

How We Built It

We designed CareLink+ with modular AI architecture, focusing on interoperability and explainability:

Frontend: Next.js + Tailwind + Framer Motion for sleek, glassy dashboards with black-and-blue accents.

Backend: Supabase for authentication, PostgreSQL for data persistence, and RESTful APIs for real-time subscriptions.

Voice Pipeline: Vapi SDK for voice recognition, integrated with Fish Audio for diarized transcription.

AI Triage Engine: Powered by Claude, leveraging a RAG system with ChromaDB for embedding-based retrieval of discharge instructions and medication info.

Automation Layer: Composio APIs connect Google Calendar, Slack, and Gmail for intelligent care coordination.

Visualization: Elastic + Kibana for data dashboards, augmented with custom charts (Recharts) for vitals and adherence trends.

Challenges We Faced

⚙️ Real-Time Data Coordination: Managing continuous voice input streams while keeping triage results consistent across sessions.

💬 Balancing UX and Compliance: Designing a smooth conversational flow while ensuring medical accuracy and privacy.

🧩 RAG Fine-Tuning: Structuring embeddings for bilingual discharge documents and ensuring relevant retrieval from mixed-language queries.

🌐 System Integration: Getting multiple APIs (Claude, Vapi, Composio, LiveKit, Supabase) to communicate efficiently without latency spikes.

What We Learned

Building real healthcare software is as much about trust as it is about technology.

Voice-first AI must adapt to emotional tone, not just content.

Seamless coordination between AI models, automation tools, and real-time data pipelines can make digital care truly human-centered.

We left this project with a deeper appreciation for how AI + empathy can redefine patient care — one voice at a time.

Built With

• claude
• composio
• livekit
• supabase
• vapi