Quiet Check-In

Worried about a parent living alone, but not ready for cameras? Quiet Check-In reads Ring sensor timing, learns their routine, and gently lets you know when something seems off.

Comment

Inspiration

Every year, millions of adult children face the heart-wrenching "caregiver’s dilemma" with aging parents who live independently:

  1. Intrusive surveillance cameras strip seniors of dignity, privacy, and autonomy. Elders routinely unplug indoor cameras, point them at walls, or refuse them altogether.
  2. Panic buttons, smartwatches, and pendants are frequently forgotten on nightstands, left in chargers, or discarded in drawers because seniors don't want to feel "frail" or "sick."

We asked ourselves a fundamental question: What if the safest home is one where peace of mind requires zero cameras, zero microphones, and zero wearables to remember?

Everyday Ring hardware contact sensors on medicine cabinets and pantry doors, motion detectors in hallways, and video doorbells on front porches already registers the natural heartbeat of a home. We realized that "Absence is the Signal." When an elder goes about their day, the home naturally breathes. If the morning tea isn't brewed, the medicine cabinet isn't opened, or the hallway stays still past normal routine windows, absence tells the story long before a crisis strikes.

What it does

Quiet Check-In is a privacy-first ambient routine intelligence platform that transforms Ring device event metadata into an invisible, non-intrusive safety net for seniors and their families.

  • 🚫 Zero Video Streams & Zero Audio Recordings: Operates entirely on cryptographic event timestamps and device labels (e.g., Door Contact Opened: 08:14 AM).
  • 🧠 Statistical Routine Intelligence (KDE + MAD): Instead of rigid alarm clocks, our engine uses Gaussian Kernel Density Estimation (KDE) to learn an individual senior's organic routine anchors (morning wake-up, breakfast, medication, evening wind-down). It dynamically calculates tolerance buffers using Median Absolute Deviation (MAD): $$\text{Tolerance Window} = \text{Median} \pm \max(45\text{m},\ 2.5 \times 1.4826 \times \text{MAD}) + 30\text{m grace}$$
  • 🪜 Deterministic Escalation Ladder:
    • L0 (All Quiet & Well): Normal routine observed. Green serene status.
    • L1 (Gentle In-App Check): Routine drift detected past grace window. Logged quietly in the family dashboard.
    • L2 (Direct Family Notification): 60 minutes past grace window. Amazon SES dispatches a calm, one-tap check-in email to the primary caregiver.
    • L3 (Circle Broadcast): 120 minutes without acknowledgment. Promotes to the wider family circle.
  • 🤖 Amazon Bedrock AI Narrator with Clinical Guardrails: Powered by Claude on Amazon Bedrock Converse API, Quiet Check-In translates raw sensor signals into warm, objective, non-alarmist summaries. Strict guardrails filter out medical diagnoses ("stroke", "heart attack") and panic words, backed by a deterministic template fallback.
  • 💬 Strands AI Family Assistant: Caregivers can open an interactive chat drawer to ask natural questions like: "Did Mom take her morning pills?" or "Why was an alert sent at 10:30 AM?" Grounded tools inspect live DynamoDB signals to provide factual answers.
  • 🎮 Interactive Time-Travel Console: A live simulation toolbar embedded directly in the dashboard allows reviewers, doctors, and family members to test scenarios (Normal Morning, Missed Medicine Routine, Away Mode, Sensor Battery Offline) with one click.

How we built it

We designed Quiet Check-In as an end-to-end, enterprise-ready serverless system across the full stack:

1. Ring Developer API Integration & Security

  • Ingests Ring Developer API v1.1 Webhooks over Amazon API Gateway HTTP API.
  • Performs constant-time HMAC-SHA256 signature verification (X-Ring-Signature) over unparsed raw request bytes to prevent timing attacks.
  • Enforces strict 300-second timestamp drift tolerance to prevent replay attacks and leverages meta.request_id in DynamoDB with 48-hour automated TTL for idempotent deduplication.

2. Open-Source Mini-Challenge: ring-partner-kit

We abstracted our integration into an open-source, MIT-licensed Python package located in ring-partner-kit/:

  • RingWebhookVerifier: Production-grade HMAC-SHA256 verifier.
  • RingApiClient: Complete OAuth2 PKCE lifecycle client with exponential backoff and rate-limit tracking.
  • RingTwinSimulator: Hardware twin CLI (ring-twin) that generates synthetic Poisson/Gaussian routine signals for automated local testing.

3. AWS Serverless Architecture (AWS SAM)

Defined completely in template.yaml for 1-click cloud deployment:

  • AWS Lambda (Python 3.12 / Arm64): Lightweight, micro-second execution using mangum.
  • Amazon DynamoDB (Single-Table Design): Stores circles, device baselines, active alerts, and deduplication keys with sub-10ms queries.
  • Amazon EventBridge Scheduler: Runs a 5-minute recurring cron trigger to evaluate routine anchors and advance the escalation ladder.
  • Amazon SES: Delivers responsive, one-tap action emails with authenticated caregiver feedback tokens.
  • Amazon Bedrock: Leverages Claude Haiku / Sonnet via the Converse API with tool-use schemas.

4. Modern Web Dashboard

  • Frontend: Built with Next.js 14, React, Tailwind CSS, and Lucide icons.
  • Includes dedicated tabs for live daily timelines, learned anchor distributions, personalized care circle escalation contacts, and an interactive Time-Travel dock.

Challenges we ran into

  1. The Raw-Body Webhook Verification Catch-22: Modern web frameworks (FastAPI/Starlette) and API Gateways automatically parse JSON bodies. However, cryptographic HMAC-SHA256 verification fails if keys are reordered, whitespace changes, or character encodings shift. We resolved this by preserving unconsumed raw byte buffers at the ASGI middleware level and verifying before deserialization.
  2. Preventing Alarm Fatigue: Elders aren't factory machines they wake up earlier on Tuesdays for gardening, sleep in on Sundays, or take afternoon naps. Rigid static timers caused unacceptable false alarms. By implementing Median Absolute Deviation (MAD) over Gaussian KDE clusters, our system dynamically widens tolerance bands to accommodate human life while catching genuine anomalies.
  3. Sensor Disconnections vs. Human Emergencies: What happens if a Ring sensor battery dies or Wi-Fi drops? A naive system triggers a terrifying emergency alert. We built an offline-state filter: if a device reports low battery or disconnects, Quiet Check-In marks the device as impaired and suppresses inactivity alerts, notifying the caregiver to replace the battery instead.
  4. Bedrock AI Guardrails & Hallucination Mitigation: In elder care, an AI saying "Mom might be incapacitated" creates panic. We integrated custom Bedrock Guardrails that strictly forbid medical assumptions, emergency claims, and clinical diagnosis language. If Bedrock exceeds token limits or throttles, a deterministic template narrator takes over instantly with zero downtime.

Accomplishments that we're proud of

  • Empirical Benchmark Performance: Evaluated across 150 simulated person-days with 5 distinct senior personas (early riser, strict medication taker, irregular wanderer, nocturnal reader):
    • 100.0% Detection Recall on injected routine breaks.
    • 0.0 False Alarms per month during normal living.
    • 10.0 minute mean detection latency past grace windows.
  • Rock-Solid Test Coverage: Built a comprehensive suite of 47 unit and integration tests passing with 100% green status across the verifier, baseline math, escalation state machine, and API endpoints.
  • Dual Mini-Challenge Submissions:
    • AWS Builder Mini-Challenge: Full AWS SAM Infrastructure as Code (API Gateway, DynamoDB, Lambda, EventBridge, Bedrock, SES).
    • Open Source Mini-Challenge: MIT-licensed ring-partner-kit published directly in the repository with standalone CLI tools.
  • Caregiver Feedback Loop: Caregivers can tap "Mom is fine (slept in)" on any alert, which automatically widens the routine's tolerance by +15 minutes for 7 days teaching the system to respect life's changing patterns.

What we learned

  • Dignity is a Technical Requirement: Surveillance is often the laziest engineering solution to monitoring. Designing around metadata requires more sophisticated statistical modeling, but it results in a product seniors willingly accept into their homes.
  • Calm Technology Principles: Notifications should never shout when they can whisper. Moving from binary alarms to a gentle 4-step escalation ladder dramatically reduces caregiver anxiety.
  • AWS Serverless Agility: AWS SAM combined with DynamoDB single-table design and EventBridge allowed us to build an enterprise-tier healthcare-adjacent architecture with zero idle server costs.

What's next for Quiet Check-In

  • Ring App Store Launch: Complete official Ring Developer App Store certification (our review documentation is ready in RING_CERTIFICATION.md).
  • Ambient Alexa Voice Nudge: Before notifying family at Level 1, prompt an ambient Echo device in the elder's room: "Good morning Eleanor, would you like me to turn on the news?" A simple verbal response instantly satisfies the morning routine.
  • Multi-Resident Household Modeling: Expand Gaussian mixture models to separate multi-person households and pet motion from the primary elder's routines.
  • Longitudinal Wellness Drift: Detect subtle mobility trends over 6-month horizons (e.g., gradual 15-minute slowing in morning hallway transit or increased nighttime bathroom visits) to provide proactive insights for geriatric care planning.

Built With

Share this project:

Updates

Submission history