CareLoop: Ambient Care Companion

An agentic Alexa + Fire TV companion that learns daily routines, notices when something's off, and gently escalates to family, so elders stay independent and families stay reassured.

Comment

Inspiration

Millions of families worry about an older parent or grandparent living alone. Cameras feel invasive, and wearables get forgotten or refused. We wanted care that works quietly, using signals from devices people already have at home, and that respects their dignity and privacy.

What it does

CareLoop is a privacy-first care agent for families. It watches simple daily-routine signals (door, motion, kitchen, Alexa, TV and medication confirmations) and decides whether to wait, nudge, or escalate.

  • Explainable decisions: every alert comes with a plain-language reason, such as "No activity for 210 minutes during waking hours."
  • Four-step escalation ladder: watching quietly, then an Alexa voice nudge, then a large-text Fire TV alert, then a message to family. It only moves up if the earlier step gets no response.
  • Quiet hours: no alerts are raised between 10 PM and 7 AM, so a sleeping relative is never flagged.
  • Calm family dashboard: a status ring, the escalation ladder, alert history and recent activity, all in high-contrast, large, readable type.
  • Answering clears alerts: any sign of life (talking to Alexa, motion, taking medication) marks open alerts as answered.

How we built it

  • Backend: Node.js and Express with a MySQL database for users, events and alerts.
  • Care agent: an explainable rules engine that looks at time since last activity, medication confirmation and time of day. It is designed so that Amazon Bedrock can replace the rules for richer reasoning.
  • Frontend: a custom HTML, CSS and JavaScript dashboard with no heavy framework, so it loads fast on a TV or tablet.
  • Security: bcrypt password hashing, JWT sessions in httpOnly SameSite cookies, fully parameterized SQL, Helmet with a strict Content Security Policy, rate limiting, input validation and per-user data isolation.

Challenges we ran into

  • Avoiding false alarms. Waiting is a decision too, so the agent only escalates after a sustained gap.
  • Designing for older users: large text, a calm tone, and no panic colors until they are needed.
  • Locking the app down so one family's data can never leak to another.

Accomplishments that we're proud of

  • A working end-to-end escalation ladder with a reason for every decision.
  • A security-first backend, not an afterthought.
  • A dashboard that feels calm rather than clinical.

What we learned

Good care technology is mostly about restraint. Knowing when not to alert matters as much as detecting a problem.

Honest note on scope

In this version, activity signals are logged manually or generated by built-in demo scenarios ("Normal day" and "Missed routine"). The Alexa, Fire TV and SMS steps are represented in the escalation logic and the dashboard, and the live integrations are the next step.

What's next

  • Live Alexa Skills Kit check-ins ("Good morning, did you take your tablets?")
  • A native Fire TV app for the alerts
  • Amazon Bedrock for deeper reasoning, plus Amazon SNS for family texts and calls
  • Per-person routine learning and multi-language support
  • Pilot testing with real families and care professionals

Built With

Share this project:

Updates

Submission history