Posturely 2.0

Inspiration

Posture pain is one of the biggest silent problems for people who sit for long hours — especially students, developers, designers, and office workers. I personally dealt with neck pain and shoulder strain, and realized that we don’t notice we’re slouching until it’s too late. I had previously built Posturely once before, but back then posture detection, body scans, and exercise correction required expensive servers and GPU processing. Keeping it free was impossible. When I discovered the ARM Hackathon, a single thought changed everything: What if Posturely could run entirely on-device? No servers, no cloud, no latency, no privacy issues — just pure ARM-powered intelligence.

What it does

Posturely is an on-device AI posture coach that lives on your phone. It offers: Live posture tracking using the phone camera Real-time slouch detection with subtle correction cues AI posture scoring analyzing head tilt, shoulder angle, and spine alignment Guided exercises including neck stretches and meditation sessions that monitor posture throughout Full-body posture scans processed instantly on-device Streaks, habit tracking, reminders, and a posture calendar Everything runs offline, powered by ARM-based ML acceleration. Nothing leaves the device, ensuring complete privacy.

How we built it

Mediapipe pose + custom pose logic for 2D landmark tracking On-device ML pipeline optimized for ARM processors Real-time angle math for spine, shoulder, and neck alignment Audio haptics & microfeedback for instant posture correction Offline storage for posture logs, streaks, and daily scores No servers at all — zero cloud/zero backend architecture The biggest breakthrough was moving from a server-dependent pose engine to a fully on-device ML workflow.

Challenges we ran into

Rewriting the entire system to remove cloud inference Optimizing posture detection to avoid overheating on older Phones Achieving stable real-time landmark tracking under varying lighting conditions Building a meditation mode that tracks posture continuously and quietly Creating accurate angle calculations for form correction Making the app fully functional without user frustration or noise fatigue And most importantly — balancing accuracy, battery usage, and performance entirely on device.

Accomplishments that we're proud of

Posturely is now 100% on-device, with zero server dependency The entire app is free, thanks to ARM-powered ML removing cloud costs iOS version approved and live on the App Store Real-time posture scoring with almost no latency Stable tracking even on older phone models A fully private posture engine — no user data ever leaves the phone Transforming a previously expensive app into a completely offline, real-time AI posture coach

What we learned

On-device ML is not just faster — it makes apps financially viable ARM optimization drastically reduces latency and battery consumption Good UX for posture correction must be gentle, not spammy Angle-based posture detection is surprisingly nuanced Removing cloud infrastructure simplifies everything — support, cost, and privacy You don’t need big servers to deliver powerful AI experiences

What's next for Posturely 2.0

Android version (after stabilizing pose detection on older devices) New AI exercises with rep counting and form correction Full-body posture heatmaps Long-term posture analytics and insights A “silent mode” using only subtle haptics Advanced ergonomic recommendations Widgets and lock-screen posture reminders The goal is simple: Make healthy posture effortless — for everyone, everywhere, on any device.

Built With

• arm
• kotlin
• mediapipe
• multiplatform
• pose