CareLoop: Voice Care Companion for Elders

Inspiration

Millions of elders live alone, and their families worry every day: did she take her medicine, did she eat, is she lonely? Most care apps need typing, small buttons and patience. We wanted something that works by simply talking.

What it does

CareLoop is a voice-first care companion. The person just speaks: "I took my Metformin", "I skipped lunch", "I'm feeling lonely today". CareLoop logs it, replies in a warm voice in her own language, and watches her routine. When something looks wrong, like a missed medicine or a low-mood trend, it alerts the family on a calm dashboard. It supports English and Hindi, with spoken replies and live subtitles.

How we built it

• A self-hosted MCP server (Streamable HTTP, spec 2025-11-25) exposes six care tools: log_medication, daily_checkin, get_wellness_summary, detect_anomalies, alert_family and get_care_plan. This is the integration pattern Alexa+ uses.
• Amazon Polly neural voices with a browser-voice fallback, plus the Web Speech API for the mic with live captions.
• Next.js and TypeScript, Drizzle ORM, and a Tailwind UI with a calm cream-and-green design.
• A "Run demo scenario" button replays a missed-medicine day end to end.

Challenges we ran into

Making voice input and output take turns without getting stuck, keeping the reply language in sync with the selected language, and making the replies short and simple enough for elders to understand.

What we learned

MCP makes tools easy to plug into any voice assistant, and a calm design with short spoken sentences matters more than features.

What's next for CareLoop

Real Alexa+ device integration, caregiver mobile notifications, and more languages.

Note: CareLoop uses seeded demo data for one fictional person. It is not medical advice. It only logs, summarizes and alerts family.

Built With

• amazon
• api
• context
• css
• drizzle
• model
• next.js
• node.js
• orm
• polly
• protocol
• speech
• tailwind
• typescript
• vercel
• web