Inspiration
The idea for Rakshak started with something personal: thinking about my grandmother and how vulnerable recovery can be when an elderly patient returns home after surgery.
After discharge, recovery doesn't stop. Patients still need to manage pain, mobility and symptoms, while families and doctors may have limited visibility until the next follow-up. For an elderly patient, even a simple fall can turn a normal recovery into an emergency.
That made me think: what if a patient could simply wear a smartwatch and have an intelligent safety net throughout their recovery?
What it does
- Voice-based daily check-ins for symptoms and recovery updates.
- AI-powered conversations that ask contextual follow-up questions.
- Wearable monitoring of heart rate, steps and SpO₂.
- Personalized recovery tracking by comparing current progress with the expected recovery curve.
- Green / Amber / Red risk assessment to identify patients who may need attention.
- Fall detection and SOS using the smartwatch's accelerometer and gyroscope.
- Real-time clinician dashboard with patient trends, risk levels and alerts.
- Discharge PDF processing to extract relevant recovery instructions and medical context.
How I built it
I built Rakshak as a full-stack system consisting of:
- Next.js for the patient, clinician and emergency interfaces.
- Node.js, Express and TypeScript for the backend.
- WebSockets for real-time events and dashboard updates.
- Sarvam Saaras v3 for speech-to-text.
- Sarvam 105B for conversational recovery assessment.
- Sarvam Bulbul v3 for voice responses.
- Gemini for emergency triage.
- Wear OS with Kotlin and Compose for smartwatch integration.
- PostgreSQL / Supabase for patient data and recovery history.
- Twilio for emergency SMS, WhatsApp and voice alerts.
Challenges I ran into
- Making AI conversations feel natural while still collecting clinically useful information.
- Combining subjective symptoms with wearable signals.
- Designing reliable fall detection using smartwatch sensors.
- Handling real-time communication between patients, doctors and emergency workflows.
- Reducing false alerts while ensuring important events are not missed.
- Designing the system around clinician oversight rather than fully autonomous medical decisions.
Accomplishments I'm proud of
- Built a working end-to-end post-discharge monitoring prototype.
- Integrated a real Wear OS companion with health signals and fall detection.
- Implemented voice-based AI check-ins instead of relying only on forms.
- Built a live doctor dashboard with recovery trends and risk assessment.
- Connected the complete path from patient → AI → risk assessment → clinician → emergency escalation.
- Added discharge-document processing so the patient's recovery context can be personalized from the beginning.
What I learned
Building Rakshak taught me that healthcare AI is not just about having a powerful model. The real challenge is connecting AI with reliable data, hardware, real-time systems and human decision-making.
I learned how to build multimodal workflows combining voice, wearable data and structured medical information. I also learned the importance of keeping clinicians in the loop when building systems that can influence healthcare decisions.
What's next for Rakshak
The next step is to move from a working prototype toward real-world validation.
- Test Rakshak with real post-surgery recovery workflows.
- Improve AI risk assessment using clinical feedback.
- Improve fall-detection accuracy across different users and activities.
- Add more wearable signals and supported devices.
- Integrate with hospital systems and existing clinical workflows.
- Expand beyond post-surgery recovery into elderly care, chronic conditions and remote patient monitoring.
The long-term vision is simple: give every patient a digital safety net between leaving the hospital and fully recovering at home.
Log in or sign up for Devpost to join the conversation.