Student-built local AI project

SitRight turns a webcam into a private posture feedback loop.

The app detects upper-body landmarks in the browser, scores posture with a tiny custom neural network, and gives real-time feedback without accounts, uploads, or server inference.

100%
browser-side inference
7
engineered posture features
929
model parameters

What it demonstrates

A compact end-to-end AI product

01

Real-time feedback

MediaPipe runs pose detection in the browser and the UI updates posture state at roughly 10 frames per second.

02

Custom scoring model

A small dense neural network scores seven geometric posture features instead of shipping frames to an API.

03

Session analytics

Scores are charted after monitoring ends, with a CSV export for reviewing posture trends or presenting results.

How it works

From webcam landmarks to posture score

1

Camera frame

The browser requests webcam access and keeps the video stream local.

2

Pose landmarks

MediaPipe extracts nose, ear, and shoulder landmarks with WebGL acceleration.

3

Feature engineering

Distances, ratios, and angles describe head position and shoulder alignment.

4

Local score

A hardcoded JavaScript model returns a 0-100 posture score and status color.

Privacy and constraints

Built honestly, with visible tradeoffs

Privacy-first by architecture

Inference happens inside the page. The app does not require an account or upload webcam frames.

Best with front-facing webcams

The model was trained around a typical desk camera angle. Low light, occlusions, and steep laptop angles can reduce accuracy.

Student project scope

This is an educational AI demo, not medical advice or a clinical posture assessment tool.