SafeStance - Industrial Safety Posture Monitoring

• 

  SafeStance detects posture issues

• 

  SafeStance detects posture issue

Inspiration

I partnered with a hardware manufacturer, and spending time on their factory floor is where this idea came from. I kept noticing assembly line workers hunched over in uncomfortable positions for hours at a time. Nobody was doing anything wrong, it was just how the work happened, but you could see the strain building up. I realized these are exactly the small, repeated bad postures that turn into real injuries over months, and that most plants have no way to catch them until it's too late.

What it does

SafeStance is an industrial health and posture monitoring solution that uses computer vision to watch how workers move and flags risky posture in real time, so teams can fix problems before they turn into injuries. No wearables, no new sensors to install, just a camera.

How I built it

I used MediaPipe for pose estimation in the browser, pulling body landmarks straight from a webcam feed. The app logic is plain JavaScript, and I leaned on Codex heavily to move fast, scaffolding the pose logic, the scoring rules, and the UI. From the landmarks I compute joint angles and turn them into a posture risk score, then surface it live on a simple dashboard. The whole thing is a static front end, so I deployed it on Netlify.

Challenges I ran into

Keeping detection accurate in real time without lagging the browser took real tuning. Lighting and partial views threw off the landmarks more than I expected, and deciding what actually counts as "risky" posture turned out to be an ergonomics question, not just a coding one. The hardest part was turning noisy landmark coordinates into a score a human could actually trust.

Accomplishments that I'm proud of

I shipped a working, deployed product that runs live in the browser, entirely on hardware people already own. Starting from a real problem I saw on an actual factory floor and ending with a link someone can open and try feels like the whole point of a hackathon.

What I learned

Getting a pose model to run is the easy part. The real work is smoothing jittery data, picking sensible angle thresholds, and understanding how much camera placement changes everything. I also got faster at working alongside Codex, learning when to trust its output and when to rewrite it myself.

What's next for SafeStance - Industrial Safety Posture Monitoring

Aligning the scoring with established standards like REBA and RULA, supporting multiple people in frame, and adding trend logging over time so safety teams can see risk patterns, not just single moments.

Built With

• chatgpt
• codex
• javascript
• mediapipe
• netlify