LifeLink 360: A Patient Practice Partner for a Friend Learning a New Language

A developer built LifeLink 360, a browser-based assistive communication web app for patients with severe motor impairments such as ALS or locked-in syndrome, using MediaPipe Face Mesh to detect 16 eye-blink patterns from a standard webcam. The app sends recognized blink patterns to the Backboard.io API, where GPT-4o generates contextual responses spoken aloud via the Web Speech API, with a Node.js/Express backend hosted on DigitalOcean's App Platform. The developer says the architecture is ready to scale to a DigitalOcean GPU Droplet for private, HIPAA-compliant local inference with open-weight models like Llama 3.

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend https://dev.to/challenges/hacktoberfest-weekend-2026-10-01 LifeLink 360 is an assistive communication web app designed for patients with severe motor impairments such as ALS, locked-in syndrome, or post-stroke recovery . I built this for my friend Friend's Name , who is a patient care partner. We realized that patients who cannot speak or move their hands often struggle to communicate basic needs like "I need water," "I am in pain," or "Yes/No." Traditional eye-tracking hardware is expensive, bulky, and inaccessible to many families. LifeLink 360 solves this by turning any standard laptop or tablet webcam into a powerful communication device. Using 16 specific eye-blink patterns single, double, triple, long blink , the patient can trigger pre-set phrases. The app then uses AI to generate a compassionate, contextual response and speaks it out loud using the Web Speech API. It gives the patient a voice, and it gives the caregiver peace of mind. This project was built using a modern, open-source AI stack combined with cloud infrastructure: Computer Vision Frontend : I used MediaPipe Face Mesh an open-source framework from Google running directly in the browser. It tracks 468 facial landmarks in real-time. I calculated the Eye Aspect Ratio EAR to determine if the eye is open or closed, and used timing logic to distinguish between short blinks, long blinks, and gaze direction. AI Agent Backboard.io : I integrated the Backboard.io API as the "brain" of the application. When a patient blinks a pattern e.g., 3 blinks = "Yes, please" , the frontend sends this to Backboard. The LLM GPT-4o processes the request using a specialized System Prompt designed for empathy and brevity, and returns a natural language response. Backend & Hosting DigitalOcean : I built a lightweight Node.js/Express backend to securely handle API keys and serve the static frontend. I deployed the entire application on the DigitalOcean App Platform, which provides automatic HTTPS, global CDN distribution, and seamless scaling. The backend is currently running on a basic Droplet, but the architecture is ready to scale to a DigitalOcean GPU Droplet to run open-weight models like Llama 3 locally for 100% private, HIPAA-compliant inference. Audio Output: The app uses the browser's native SpeechSynthesis API to read the AI's response out loud, ensuring the patient and caregiver can both hear the communication. Open innovation is the only reason this project exists. Accessibility: Closed, proprietary eye-tracking hardware can cost thousands of dollars. By using MediaPipe open-source and standard webcams, I made this tool accessible to anyone with a laptop. Customization: Because we used open frameworks, a caregiver can easily tweak the BLINK THRESHOLD in the code to match the specific eye shape and fatigue level of their patient. You cannot do that with a black-box medical device. Privacy & Local Inference: The future of this project involves deploying open-weight models on a DigitalOcean GPU Droplet. This means patient data never has to leave the home. Open innovation allows us to build medical tools that are both intelligent and private. I built this primarily using Backboard.io's API We are entering the following partner categories: Best Use of DigitalOcean: For hosting the Node.js backend on App Platform and preparing the architecture for GPU Droplet inference to run open-weight models privately. Best Use of Backboard.io: For using their API to handle the contextual, patient language tutor layer that powers the conversation. Team Submissions: This project was built by: