KinGuard: Ambient Cognitive Care Protocol

Inspiration

Over 55 million seniors worldwide live independently while coping with early-stage cognitive impairment, memory lapses, or dementia. Daily life is shadowed by acute anxiety: unexpected visitors at the front door, repetitive medication uncertainty loops, and standard voice assistants that confront or argue when asked the same question twice. In clinical memory care, arguing directly violates Validation Therapy—it triggers agitation and worsens cognitive decline. We built KinGuard to turn the home into an ambient, empathetic cognitive protector.

What It Does

KinGuard is a zero-friction, voice-first ambient cognitive care system powered by Alexa+, Ring smart doorbells, and Amazon Bedrock:

1. Context-Aware Doorbell Escort (Ring + Alexa+ Bridge): When a visitor rings, Martha (78) asks: "Alexa, who is at the front door?" KinGuard calls the MCP tool kinguard_get_door_snapshot, identifies the verified Amazon courier, reassures Martha that the door is locked, and offers to speak to them. Upon voice confirmation, it broadcasts soothing two-way synthesized audio directly through the Ring porch speaker.
2. Ambient Medication Grounding: Martha asks: "Did I take my blood pressure pills today?" KinGuard invokes kinguard_verify_medication_state to inspect the smart pill tray sensor and provides warm, unambiguous verification.
3. Clinical Validation Therapy on Amazon Bedrock: KinGuard routes ambient telemetry through Amazon Bedrock (Claude 3.5 Sonnet / Haiku) using specialized clinical validation prompts—never invalidating emotions, always providing calm 1–2 sentence vocal dialog.
4. Caregiver Escalation via Amazon SNS & EventBridge: High anxiety, panic loops, or security risks trigger kinguard_escalate_caregiver, dispatching an instant SMS summary to family caregivers and logging an immutable audit event to Amazon EventBridge.

How We Built It

• Alexa+ Self-Hosted MCP Server (Spec 2025-11-25): Python, FastAPI, and Uvicorn exposing JSON-RPC 2.0 endpoints over Streamable HTTP (/mcp) and Server-Sent Events (/mcp/sse), declaring 5 specialized cognitive tools.
• Amazon Bedrock Integration: Boto3 runtime connecting to Claude 3.5 Sonnet & Haiku with clinical validation guardrails and resilient local fallback.
• Amazon SNS & EventBridge: CloudFormation stack (cloudformation_stack.yaml) orchestrating emergency caregiver dispatches.
• Interactive Live Simulator: High-fidelity web app with live Ring video canvas, Web Speech API voice synthesis & microphone recognition, dynamic sensor toggles, and real-time MCP wire terminal.

Challenges We Ran Into

• Fine-tuning non-blocking Server-Sent Events (SSE) buffering behind HTTP proxies.
• Sub-second vocal latency optimization using Claude 3 Haiku for near-instantaneous vocal responses.
• Developing a full-fidelity interactive canvas simulator for Ring porch telemetry.

Accomplishments We're Proud Of

• 100% real, dynamic execution—zero static mockups.
• Native compliance with the cutting-edge Model Context Protocol (2025-11-25) Streamable HTTP spec.
• Humanitarian impact for aging-in-place seniors and family caregivers.

Built With

• alexa+
• amazon-bedrock
• amazon-eventbridge
• aws-sns
• fastapi
• model-context-protocol
• python
• ring