High School Hackathon Project

Help is already home.

ResQ reaches the seniors who live alone, off the grid — where Wi‑Fi is unreliable and reliable tech is far out of reach. Because no one should face an emergency alone.

On-Device AI Satellite, Not Wi‑Fi Privacy by Design Solar Powered Multilingual
🧠
MoveNet
AI Model (TensorFlow.js)
📍
17 Keypoints
Detected per frame
< 50 ms
Inference latency
🔒
100% On-Device
Zero data uploaded
🎯
3 States
Upright · Sitting · Lying
🌐
WebGL
GPU-accelerated backend
Interactive Demo

See ResQ in Action

TensorFlow.js MoveNet runs entirely on your device — nothing is uploaded. Alerts, ETA, and dispatch are simulated for this demo.

📷
Camera not active

Press "Start Demo" to enable your camera.
Everything runs on‑device — nothing is uploaded.

🔒 Vision on‑device only
Demo Controls

Allow camera access when prompted. Blocked? Demo runs in simulated mode automatically.

System Status
Idle
Start the demo to begin monitoring.
Risk Level
None
Event Log
No events yet — start the demo above.
Built for the Real World

Every Detail Matters

Designed from the ground up for people with limited or unreliable technology — inspired by a grandmother in rural Tunisia.

📿

Wearable Necklace

One press calls for help instantly — no smartphone, no app, no password. Designed for arthritic hands and low dexterity.

🧠

On-Device Fall Detection

MoveNet AI classifies posture and detects lying-down emergencies entirely on the robot — zero data leaves the home.

💬

Conversational Check-In

ResQ speaks first, giving the user 10 seconds to say "I'm OK." Escalation only happens if there's no response.

📡

Satellite, Not Wi‑Fi

Connects over satellite directly, so it works in rural areas with no broadband — the digital divide doesn't apply here.

☀️

Solar Charging

A small solar panel keeps ResQ powered indefinitely in well-lit rooms — no wall outlets, no cables to forget.

🔊

Multimodal Guidance

Audio instructions, on-screen subtitles, and visual cues work together for users with hearing or vision differences.

🔒

Privacy by Design

Camera only activates when an alert is raised — never always-on. No video is stored, streamed, or uploaded.

🌍

Multilingual Support

Guidance plays in the user's language — designed for communities where English is not a first language.

Powered by TensorFlow.js

AI Under the Hood

ResQ runs a full computer-vision pipeline locally — no cloud, no server, no latency. Here's exactly how it works.

Inference Pipeline

📷
Video Frame Capture
getUserMedia streams live webcam frames directly into a canvas element at up to 30 fps.
MediaStream API
⚙️
TensorFlow.js WebGL Backend
Each frame is converted to a tensor and fed into the GPU via the WebGL backend — no CPU bottleneck.
@tensorflow/tfjs-backend-webgl
🎯
MoveNet SinglePose Lightning
Google's lightweight pose model predicts 17 body keypoints with (x, y, confidence) in under 50 ms per frame.
@tensorflow-models/pose-detection
📐
Geometric Posture Classifier
Shoulder-to-hip vector angle determines upright vs. sitting vs. lying. Sustained lying for 3 s triggers risk escalation.
Custom heuristic · confidence threshold: 0.3
🚨
Risk State Machine
Low → Medium → High risk levels gate check-in, spoken guidance, and caregiver alert — all in the browser.
Web Speech API · no network required

Model Specs

Model
MoveNet Lightning
Keypoints
17 body landmarks
Latency
< 50 ms per frame
Backend
WebGL GPU
Framework
TensorFlow.js 4.17
Deployment
100% client-side
🔐

Privacy Guarantee

The model weights load from a CDN once. After that, every frame is processed entirely in the user's browser using GPU memory. No image, frame, or keypoint is ever sent to a server. The camera only activates when an alert is raised — never always-on.

17-Point Skeleton Map

MoveNet outputs a confidence-weighted (x, y) coordinate for each of these landmarks every frame.

nose shoulder hip knee ankle

Posture Classification Logic

🚶
Upright
Shoulder–hip vector is mostly vertical; ankles visible below knees
🪑
Sitting
Knees near hip height; ankles absent or at same level
⚠️
Lying Down
Horizontal shoulder–hip span exceeds vertical span by 1.15×
Step by Step

How ResQ Works

From necklace press to caregiver contact — under 60 seconds.

1

Necklace Press

User presses button or a fall is detected automatically.

2

Robot Moves

ResQ navigates autonomously to the user's location.

3

AI Assesses

Camera + MoveNet evaluate posture on-device in real time.

4

Check‑In

ResQ asks "Are you OK?" and waits 10 seconds.

5

Risk Level

Low, medium, or high determines the escalation path.

6

Guidance

Step-by-step audio + visual first-aid instructions delivered.

7

Contact Help

Caregiver or EMS alerted over satellite with location.

Why It Matters

The Scale of the Problem

These numbers represent real people. ResQ exists because every one of them deserves safety at home.

16.2M
U.S. adults 65+ live alone — about 28% of all seniors in the community
3M+
seniors treated in U.S. emergency departments for fall-related injuries every year
1 in 5
fall victims lie on the floor for over an hour — a "long lie" that worsens outcomes
2.6B
people globally still lack access to digital services — ResQ works without internet