Rehab Assist

Inspiration

Rehabilitation exercises are often done without proper supervision outside of clinic sessions, which can lead to poor form, slow recovery, or even re-injury. We wanted to build something that acts like a real-time digital assistant — giving patients immediate feedback while allowing physiotherapists to remotely monitor progress. Our goal was to bridge the gap between clinic visits and at-home rehab with accessible computer vision technology.

What it does

RehabAssist is a rehabilitation exercise analysis platform that uses real-time pose detection to help users perform exercises correctly.

It provides:

Live pose tracking and skeleton visualization using the camera

Real-time form scoring and feedback

Automatic rep counting for mini squats

Joint angle analysis to validate movement quality

Progress tracking and analytics over time

A physio-client management system for monitoring recovery

The system guides users through exercises while continuously analyzing motion to ensure correct technique.

How we built it

We built RehabAssist as a modern web application focused on accessibility and real-time performance.

Tech stack:

React + TypeScript for a scalable frontend architecture

MediaPipe for real-time pose estimation and landmark tracking

Supabase for data storage, client management, and analytics

Tailwind CSS + shadcn/ui for a clean, responsive UI

Vite for fast development and build tooling

We implemented a state-based exercise system (s1 → s2 → s3) that tracks movement phases. Joint angles are calculated from pose landmarks and used to validate form, score reps, and trigger feedback.

Challenges we ran into

Dealing with real-time pose data reliability. Camera angle differences, lighting conditions, and slight body variations created noisy measurements that affected form scoring.

Other challenges included:

Stabilizing pose detection to prevent rep miscounts

Connecting the Frontend to the Backend

Translating raw joint data into meaningful feedback for users

Accomplishments that we're proud of

Building a fully working real-time exercise analysis system in the browser

Successfully integrating MediaPipe with live feedback and rep counting

Creating a clean and intuitive visualization of skeletal tracking

Implementing a scalable architecture that supports multiple clients and future exercises

Turning raw computer vision outputs into actionable rehab insights

What we learned

Through building RehabAssist, we learned:

Real-time computer vision requires strong state management and filtering

Human movement analysis is highly context-dependent — hardcoded thresholds only go so far

UX matters just as much as technical accuracy in healthcare-focused tools

Structuring code for extensibility early makes adding new exercises easier later

We also learned how important personalization is in rehabilitation — no two users move exactly the same way.

What's next for Rehab Assist

The next step is making the system fully customizable for physiotherapists.

We plan to add:

Custom angle editing, allowing physios to define acceptable joint ranges

Depth customization, so therapists can specify how deep a client should squat based on recovery stage

Exercise-specific presets for different injury types

Adaptive difficulty that adjusts as a client improves

Expanded exercise support beyond mini squats

Improved analytics dashboards for long-term progress tracking

Our long-term vision is to create a platform where physiotherapists can tailor exercise parameters to each patient, enabling more personalized, safe, and effective rehabilitation at home.

Built With

• mediapipe
• radixui
• react
• supabase
• tailwindcss
• typescript
• vite