Inspiration
In early Alzheimer's, the answer doesn't stay. Rosa, 76, asks "Where's Ana?" twenty times before lunch, and each time her daughter's phone rings at work. More than 12 million Americans care for someone with dementia without pay. Written cues are known to reduce repeated questions (Bourgeois 1997), yet no product helps a family give one calm answer that is always there.
## What it does
- Ana answers once, in her own words, from her phone.
- Rosa's Fire TV shows that answer in big letters, then plays the family album.
- Alexa gives the same answer however Rosa asks: "Where's my girl?", "Is Ana home?"
- New photos from the phone or a family share link appear on the TV within seconds.
- An agent notices when her questions rise above her own normal and takes one small step from Ana's playbook.
Rosa only ever hears words Ana approved.
## How we built it Vega OS app (React Native) for Fire TV · Alexa skill + MCP server (2025-11-25) · AWS Lambda, DynamoDB, EventBridge, SNS, S3 + CloudFront · Claude on Amazon Bedrock (caregiver side only) · ActiveGraph agent · SwiftUI iPhone app and web app.
## Challenges we ran into
- The Vega Virtual Device hung on macOS 27, so we ran it in WSL2 Ubuntu with nested virtualization.
- React loaded twice in the TV bundle crashed the app on start. We found it by reading the device's crash dump.
## Accomplishments that we're proud of The whole loop runs live in the cloud: pair the TV, add a photo on the iPhone, and seconds later it's on the Vega TV. Every agent rule is backed by a published study, and no AI speaks between Rosa and her daughter's words.
## What we learned Good dementia tech is calm, repetitive and almost invisible. Alerts alone don't reduce caregiver stress.
## What's next for Lumina
- Emergency contacts on the TV and in Alexa
- Alexa skill certification (request signature checks, account linking)
- A pilot with families that measures caregiver stress
Built With
- activegraph
- alexa
- alexa-skills-kit
- amazon-bedrock
- amazon-cloudfront
- amazon-dynamodb
- amazon-eventbridge
- amazon-sns
- amazon-web-services
- aws-lambda
- claude
- fire-tv
- hono
- model-context-protocol
- node.js
- python
- react
- react-native
- swift
- swiftui
- terraform
- typescript
- vega-os
Log in or sign up for Devpost to join the conversation.