RingGuard AI — Autonomous Caretaking for Ring

AI caretaking co-pilot for Ring doorbells using AWS Bedrock & Scrapling. Auto-unlocks for caregivers, verifies couriers, & escalates emergencies if a vulnerable resident needs help.

Comment

RingGuard AI 🛡️

Autonomous AI Caretaking & Access Co-Pilot for Ring Doorbells and Smart Cameras

Track: Ring (Access Control & Smart Home Automation)
Stacked Mini-Challenges: AWS Builder + Open Source
Bonus Target: Up to 10% Developer Friction Log & Tool Feedback Bonus

💡 Inspiration

Over 54 million family members in North America care for an aging parent or relative living independently. Today, standard smart doorbells like Ring provide raw, uncontextualized video and noisy push notifications ("Motion detected at Front Door").

When an elderly parent with cognitive decline or mobility issues wanders near the front entrance, or when an authorized visiting nurse arrives during a scheduled appointment, traditional doorbells suffer from two lethal shortcomings:

  1. Alert Fatigue: Caregivers receive dozens of notifications daily for passing cars, neighbors, and pets, causing them to mute or ignore urgent alerts.
  2. Critical Response Gap: If an 82-year-old resident leaves the front door ajar or becomes unresponsive, passive cameras do nothing more than send another silent notification.

We asked: What if your Ring doorbell wasn't just a passive lens, but an autonomous, explainable caretaking agent that could see, understand, verify, and act?

🚀 What It Does

RingGuard AI converts raw Ring device events into an autonomous, closed-loop safety pipeline: See → Understand → Decide → Act → Log.

  • 🔓 Autonomous Caregiver Verification & Deadbolt Unlocking: When a verified visiting nurse or physical therapist arrives during their authorized caretaking schedule, Amazon Bedrock cross-references their face and credentials, logs explainable reasoning, and auto-unlocks the smart deadbolt.
  • 📦 Scrapling Package Intelligence & Amazon Key: Ingests package tracking manifests (e.g. TBA309924810239) using Scrapling v0.4.15 anti-bot intelligence, recognizes Amazon delivery couriers, logs package drops, and offers 1-click garage/entry access.
  • 🚨 Life-Safety Emergency Escalation Watchdog: If an aging resident is detected at an open door, unsteady, or unresponsive, RingGuard immediately sounds a chime, initiates a check-in prompt, and activates a high-urgency 120-second escalation countdown. If unacknowledged, it auto-dispatches SMS/Push alerts to emergency contacts via Amazon SNS.
  • 🎙️ Alexa+ Voice Command Bar: Caregivers and residents can issue natural-language safety audits ("Alexa, who visited mom this afternoon?" or "Alexa, lock the front deadbolt and arm night watch").
  • 🎛️ Mission Control Dashboard: An Amazon-styled, safety-grade command center with multi-camera switching (Front Doorbell Pro 4, Backyard Floodlight, Echo Show 15), real-time SVG waveform audio visualizers, computer-vision bounding boxes, and an immutable audit timeline.

🛠️ How We Built It

RingGuard AI is engineered as an end-to-end, production-grade cloud and IoT architecture:

  1. Ring Hardware Integration Layer (ring-integration/):

    • Imports the authentic ring-client-api SDK for hardware token authentication, live SIP session management, and camera event streaming.
    • Includes a deterministic, high-fidelity Ring device simulator generating exact REST/SIP event envelopes for repeatable benchmark testing.

  2. AWS Bedrock Reasoning Core (agent/bedrockAgent.ts):

    • Powered by Claude 3.5 Sonnet and Amazon Nova Pro foundation models on Amazon Bedrock.
    • Equipped with agent tools (check_known_visitors, get_caretaking_config) that inspect schedules, authorized caregiver rosters, and resident caretaking profiles.
    • Enforces strict JSON output schemas: { visitor_type, confidence, action, reasoning, requires_escalation }.

  3. Amazon DynamoDB Event Store (agent/dynamoStore.ts):

    • Stores immutable audit logs across three tables: RingGuard-Events, RingGuard-KnownVisitors, and RingGuard-CaretakingConfig.
    • Single-digit millisecond query performance for real-time dashboard updates.

  4. Package Intelligence (services/scrapling_delivery_service.py):

    • Python microservice utilizing Scrapling 0.4.15 to parse Amazon delivery dispatch manifests and cross-verify courier presence at the doorstep.

  5. Safety-Grade Dashboard (dashboard/):

    • Built with Next.js 15 App Router, TypeScript, Tailwind CSS, and Framer Motion physics.
    • Adheres to the "Mission Control Obsidian" design system with authentic Amazon navigation palette (#131921, #232F3E, #FF9900, #00A8E1), fluid spring physics, and strict zero-gravity UI aesthetics.

🧗 Challenges We Ran Into

  • Ring SDK System Metrics Bundling: ring-client-api pulled transitive desktop system monitoring libraries (osx-temperature-sensor) that triggered Webpack bundling warnings in Next.js Server Components. We resolved this cleanly by decoupling the backend integration modules under a dedicated @core path alias with explicit Next.js output file tracing.
  • Strict Structured Outputs in Bedrock: Early foundation model invocations occasionally returned conversational preambles ("Here is the JSON:") before the payload, threatening automated lock actuators. We implemented robust regex-based JSON envelope extraction and schema verification fallbacks to guarantee 100% actuator safety.
  • Balancing Security vs. Autonomy: Caregivers must never be locked out, yet strangers must never be mistakenly granted entry. We solved this with dual-factor policy checks: visual identity + scheduled care window + known caregiver whitelist.

🏆 Accomplishments We're Proud Of

  • 100% Automated Test Suite Passing: 29 out of 29 automated end-to-end and unit tests passing across normalization, Bedrock reasoning, DynamoDB storage, and action routing.
  • Real SDK Integration: Built with genuine ring-client-api and @aws-sdk/client-bedrock-runtime imports, not static mocks.
  • Under-3-Second Triage: Complete round-trip from doorbell ding to Bedrock analysis, smart deadbolt actuation, and dashboard dispatch executes in under 2.4 seconds.
  • Empirical Developer Friction Logs: Authored 4 comprehensive friction logs and 5 Amazon feature requests in docs/ to help improve the Ring and AWS developer ecosystems.

📚 What We Learned

  • How to orchestrate agentic tool calling in AWS Bedrock using both Anthropic Claude 3.5 Sonnet and Amazon Nova models.
  • The power of combining IoT edge streaming protocols (Ring SIP/REST) with serverless cloud AI for high-stakes healthcare and elder-care applications.
  • How to design safety-critical UI where visual hierarchy and latency can literally protect human life.

🔮 What's Next for RingGuard AI

  • Native Two-Way Audio Voice Agent: Bridging AWS Bedrock voice agents directly into Ring Doorbell speakers over WebRTC to allow autonomous, polite conversations with delivery drivers and visitors.
  • AWS Greengrass Edge Deployment: Deploying quantized vision reasoning models directly onto Amazon Ring Alarm Pro base stations to maintain local caregiver auto-unlock even during ISP broadband outages.
  • Wearable Fall Detection Sync: Pairing Ring indoor cameras with health wearables to correlate doorstep motion with vital signs.

Built With

Share this project:

Updates

Submission history