Signal Door

Inspiration Over 430 million people worldwide live with disabling hearing loss. Conventional smart doorbells rely on audible chimes, easily missed push notifications, and audio-only intercoms, leaving deaf and hard-of-hearing residents vulnerable to missed deliveries and communication barriers.

We built SignalDoor to transform the traditional doorbell into an accessible, visual, two-way communication hub that helps residents see who's at the door and respond without relying on sound. What It Does

Immediate Visual Alerts (< 2 seconds): Displays a persistent, high-contrast alert with a 19.3:1 contrast ratio and a low-frequency visual strobe operating at 1.25 Hz. Alerts remain visible until acknowledged rather than disappearing automatically. The interface also supports reduced-motion preferences.
AI-Powered Visitor Summaries (< 5 seconds): Uses Amazon Nova Lite through Amazon Bedrock's Converse API to analyze a visitor's camera snapshot and generate concise, plain-language descriptions, such as "A person wearing a blue shirt and holding a box." The system avoids inferring sensitive demographic attributes.
Accessible Two-Way Quick Replies (< 1.5 seconds): Provides large, easy-to-tap buttons (at least 48 × 48 CSS pixels) for responses such as "Leave it at the door," "Coming in 2 minutes," and "I can't hear you; please text me."
Exterior Visitor Screen (/visitor): Displays residents' responses outside in bold, high-contrast, 40-pixel text. An anti-stale reset clears previous instructions when a new visitor rings, preventing outdated messages from being displayed. How We Built It
Ring Webhook and REST Client: An Express endpoint verifies webhook signatures using HMAC-SHA256 and constant-time comparison with crypto.timingSafeEqual. A resilient Ring client implements exponential backoff and rate-limit handling.
Amazon Bedrock AI Integration: Sends visitor snapshots to Amazon Nova Lite through the Converse API using in-memory image data represented as a Uint8Array. The privacy-focused design avoids writing visitor images to disk or including them in application logs.
Real-Time Communication: Uses Node.js 22 and WebSockets (ws) to synchronize visitor events and resident responses in real time. Audit events are recorded using Node.js's native node:sqlite module.
Accessible Frontend: Built with React 18 and Vite, featuring high-contrast visual components, reduced-motion support, and automated accessibility checks using axe-core. Challenges We Ran Into
Balancing AI Latency with Urgent Alerts: AI inference can take several seconds, but residents need immediate notification. We addressed this with a two-phase architecture: Phase 1 displays an immediate visual alert with fallback text, while Phase 2 enriches the notification with the AI-generated visitor summary. The update uses aria-live="polite" to communicate changes to supported assistive technologies without unnecessarily interrupting users.
Reliable HMAC Verification: Parsing and re-serializing JSON can change the request body and invalidate its signature. We addressed this by capturing the original raw request body and using it for signature verification. Accomplishments That We're Proud Of
Automated Testing: 100 automated tests across 12 test files, with all tests passing.
Accessibility-Focused Design: A 19.3:1 contrast ratio, a 1.25 Hz visual strobe, persistent alerts, and reduced-motion support.
Privacy-Conscious Image Processing: Visitor snapshots are processed in memory without intentionally persisting image files to disk.
Real-World AI Performance: Achieved a reported median inference latency of 2,186 milliseconds in unmocked Amazon Bedrock tests, with generated descriptions limited to 25 words and designed to avoid demographic speculation. What We Learned and What's Next Key Takeaway: Accessibility must be built into the architecture, not added as an afterthought. Persistent alerts, fast fallbacks, sensory safety, and accessible communication are essential parts of the user experience.

What's Next:

Add smartwatch haptic alerts with Wear OS and Apple Watch.
Integrate smart lighting, including Philips Hue, for ambient visual notifications.
Support customizable quick replies and personalized accessibility preferences. SignalDoor makes doorbell communication more accessible by combining immediate visual alerts, AI-powered visitor descriptions, and simple two-way responses in one unified experience.