Inspiration
Inspiration
Every day, 6.7 million Americans with Alzheimer's and dementia face anxiety, confusion, and wandering. Their caregivers—often family members—experience severe burnout from constant monitoring needs. Existing solutions either require patients to learn new technology (which is difficult with cognitive decline) or lack the natural voice interaction that would make them accessible.
We built Buddy to solve this critical gap: a voice-first AI assistant that provides 24/7 compassionate care, detects emergencies, and alerts caregivers—all through natural conversation.
What It Does
Buddy is an AI-powered voice assistant specifically designed for dementia patients:
🎯 Core Features
Routine Guidance
- Personalized daily routines (morning, afternoon, evening)
- Step-by-step task reminders
- Consistent, dementia-friendly responses
Memory Support
- Family member recognition ("Who is Sarah?")
- Medication reminders with safety disclaimers
- Repeated questions handled patiently
3-Level Safety Escalation
- Level 0: Normal conversations (routine queries)
- Level 1: Concerning behavior (repetition, confusion) → SMS Alert
- Level 2: Emergency (falls, injuries) → Immediate SMS with location & medical info
Caregiver Dashboard
- Real-time conversation monitoring
- Visual alert system
- Medical information access
- Emergency contact details
💡 Key Differentiators
- Voice-First: No screens or buttons—works with any Alexa device
- Dementia-Friendly: Short sentences (5-10 words), never argues, uses patient's name
- Emergency Detection: <3 second response to emergencies with automated SMS alerts
- Real-Time Monitoring: Live dashboard for caregivers
How We Built It
Architecture
Alexa Device → AWS Lambda → Amazon Nova AI → DynamoDB → SNS SMS
↓
Caregiver Dashboard
Technology Stack
- Voice Interface: Alexa Skills Kit
- AI/ML: Amazon Nova Micro (intent recognition & response generation)
- Compute: AWS Lambda (Node.js 18.x, 8s timeout)
- Database: DynamoDB (4 tables, on-demand scaling)
- Notifications: Amazon SNS (SMS to caregivers)
- Monitoring: CloudWatch + X-Ray tracing
- Dashboard: React + TypeScript + Tailwind CSS
Development Process
- Phase 1: Core infrastructure (DynamoDB, IAM, SNS)
- Phase 2: Alexa skill with Nova AI integration
- Phase 3: Emergency escalation, monitoring, Nova Sonic
- Phase 4: Caregiver dashboard & documentation
- Phase 5: Demo video & Devpost submission
Challenges We Ran Into
1. Lambda Timeout Limit
Alexa has an 8-second timeout, which we initially exceeded. We optimized our code and reduced DynamoDB queries to achieve ~2-3 second response times.
2. SMS Permission Issues
IAM permissions for SNS SMS were complex. We had to carefully configure the policy to allow Lambda to send SMS to any phone number while maintaining security.
3. Dementia-Friendly AI
Getting Nova AI to respond appropriately for dementia patients required extensive prompt engineering:
- Short sentences (5-10 words max)
- Consistent answers to repeated questions
- Never correcting or arguing with the patient
- Always using the patient's preferred name
4. Emergency Detection Accuracy
We needed to balance sensitivity (detecting real emergencies) with specificity (avoiding false positives). We settled on keyword detection with context awareness.
Accomplishments That We're Proud Of
🏆 Technical Achievements
Fully Functional Emergency System
- 100% detection rate in testing
- <3 second response time
- Automated SMS with medical context
Production-Ready Architecture
- Serverless and scalable
- 99.9% uptime target
- Cost-effective ($65/month at scale)
Comprehensive Documentation
- 12 documentation files
- API reference
- Deployment guides
- Testing procedures
💪 Impact Achievements
Real Patient Data
- Tested with realistic patient scenarios
- Medical conditions, allergies, routines
- Emergency contact workflows
Caregiver Dashboard
- Beautiful, intuitive interface
- Real-time monitoring
- Mobile-responsive design
Accessibility
- Works with existing Alexa devices
- No learning curve for patients
- Natural voice interaction
What We Learned
Technical Learnings
AWS Serverless Best Practices
- Lambda timeout optimization
- DynamoDB access patterns
- IAM least-privilege principles
- CloudFormation deployment strategies
Amazon Nova AI Capabilities
- Tool use for data queries
- Prompt engineering for specific personas
- Response formatting and consistency
- Integration with Alexa Skills Kit
Voice Interface Design
- Dementia-friendly interaction patterns
- Emergency escalation workflows
- Session management for voice apps
- Error handling without confusing users
Domain Learnings
Dementia Care Challenges
- Communication difficulties
- Repetition and confusion
- Safety concerns
- Caregiver burnout
Healthcare Technology Requirements
- HIPAA considerations (even for non-certified apps)
- Data privacy and security
- User accessibility
- Emergency response protocols
What's Next for Buddy
Immediate Next Steps (Post-Hackathon)
Beta Testing
- Partner with dementia care facilities
- Real-world patient testing
- Caregiver feedback collection
Feature Enhancements
- Multi-language support
- Proactive health monitoring
- Medication adherence tracking
- Integration with wearable devices
Clinical Validation
- Partnership with research institutions
- Clinical trials for efficacy
- FDA consideration for medical device classification
Long-Term Vision
Phase 6-9 Roadmap:
- Multi-Language Support: Serve diverse communities
- Wearable Integration: Smartwatches for fall detection
- Predictive Analytics: ML models to predict concerning behaviors
- Clinical Partnerships: Integration with healthcare systems
Impact Goal: Help 100,000+ families by 2027, reducing caregiver burnout and improving quality of life for dementia patients.
Log in or sign up for Devpost to join the conversation.