RealCare

An AI agent that remembers your loved ones for you, so distance never becomes forgetting.

Comment

Inspiration

I moved away from my family to another state to build a better life for myself. What I didn't expect was how quietly the people I love would start to feel far away, not because I love them any less, but because life gets busy and time slips by without you noticing.

I've gone two to four months without checking in on my best friend, someone I haven't seen in about four years. I've also found myself forgetting to call my mom, my dad, or my siblings not out of neglect, but because I get consumed by everything I'm chasing.

This became deeply personal because of my dad, who lives with dementia. He rarely reaches out to others himself, and being far away makes it even easier for time to pass without realizing how long it has been since I checked on him. I've felt that connection quietly slipping without meaning for it to.

That's where RealCare came from.

I wanted to build something that remembers the people we care about when life gets too busy for us to remember everything ourselves.

RealCare is designed to notice meaningful moments, understand the context around our relationships, and help us know when it's time to reach out. It can prepare a thoughtful check-in, remind us about important moments, and ask for our permission before taking action.

The goal isn't to replace human connection.

It's to make sure distance, busyness, and time don't quietly turn into being forgotten.

What it does

  1. Autonomous Proactive Heartbeat Daemon: The agent continuously runs in the background, analyzing days since last contact against individualized care cadences (e.g., every 4 days for an aging grandmother in memory care vs. 14 days for a busy medical resident brother).
  2. Sensory Memory Anchors (Anti-Disorientation Guardrails): Rather than asking generic or stressful questions, the agent grounds conversation in sensory comforts—such as morning chamomile tea with honey, blue hydrangeas, or beloved family pets.
  3. Transparent Reasoning & HITL Agency: Before any message is dispatched, the Strands Agent presents its reasoning to the family member ("Eleanor responds best in late mornings without open-ended quizzes; grounding in her lavender garden reduces anxiety"). The user retains full agency to edit, approve, or discard the draft.
  4. Triple-Channel Care Alerts: Caregivers receive notifications through three distinct delivery mechanisms:
  5. In-app ambient chime (Web Audio synthesizer) with sliding glassmorphic live alert cards.
  6. OS-level native desktop Web Push (fires even when the browser is minimized).
  7. Simulated SMS & WhatsApp webhooks for emergency dispatch.
  8. Shared Family View: Coordinates touchpoints across siblings, adult children, and grandchildren so care is evenly shared and no one experiences caregiver burnout.

How we built it

Strands Agents SDK (strands-agents): Powers the core autonomous loop with custom @tool get_overdue_contacts and @tool fetch_person_context implementations.

  • Amazon Bedrock & Dual-Engine Architecture: Built to run seamlessly on Amazon Bedrock Claude 3.5 Sonnet / Haiku 4.5 US inference profiles, with a zero-dependency local simulation mode for continuous evaluation and testing.
  • FastAPI & Python 3.11: REST API layer managing file-backed persistence and notification dispatch.
  • Vanilla CSS & Modern Typography: Premium glassmorphic interface utilizing Google Fonts (Outfit, Plus Jakarta Sans, Fraunces), custom Remix Icons, and mobile navigation tabs.
  • Live Deployment: Hosted on cloud infrastructure with Docker and automated CI/CD directly from GitHub.

Challenges we ran into

  • Dynamic API Base Resolution across Mobile & Desktop: Overcoming cross-origin and network resolution nuances so the frontend could seamlessly trigger backend agent routines regardless of whether it was running locally or over the live cloud deployment.
  • Designing Low-Latency Memory Retrieval: Grounding the agent in deeply nuanced, individual memory anchors without creating excessive token latency, ensuring draft suggestions feel instantaneous yet emotionally grounded.
  • Extreme Time Constraints: Designing, developing, containerizing with Docker, testing on cloud infrastructure, and producing demo media in under 24 hours before the hackathon deadline.

What we learned

More about Amazon Bedrock and how amazing it is, and also about the Strands Agents SDK. I learned that if one can dream it, one can surely create or make it happen.

Accomplishments that we're proud of

Having the project live and every function had in mine is working

What we learned

More about Amazon Bedrock and how amazing it is, and also about the Strands Agents SDK. I learned that if one can dream it, one can surely create or make it happen.

What's Next for RealCare

  • Direct integration with Twilio and WhatsApp Business APIs for two-way family group chats.
  • Smart home ambient display integration (Amazon Echo Show / Fire TV widgets) for elderly loved ones without smartphones.
  • Multi-lingual empathetic translation for cross-generational immigrant families.

Built With

Share this project:

Updates

Submission history