Movemend

Inspiration

MoveMend was born from the need for accessible, personalized physiotherapy. In the U.S., most physiotherapy outcomes rely heavily on patients completing exercises at home, yet many only attend in-person sessions once or twice a week. Without real-time guidance, it's easy to perform movements incorrectly, risking re-injury or delayed recovery. We wanted to build a tool that empowers users with intelligent, real-time feedback so they can confidently follow their exercise programs and make every rep count — even outside the clinic.

What it does

MoveMend harnesses AI-powered pose estimation to track your body's movements in real-time. Using BlazePose's advanced 3D keypoint detection technology, our system analyzes joint angles, body alignment, and movement patterns during exercises like squats and lunges. The AI then generates physiotherapist-style guidance delivered through visual and verbal cues, offering immediate feedback to help you perfect your form, maximize results, and prevent injuries.

How we built it

We built MoveMend with a React frontend and integrated TensorFlow.js using BlazePose for real-time 3D pose detection. The app processes keypoints to calculate joint angles and body positions, which are analyzed against ideal biomechanics. We use the Groq API to generate concise, physiotherapist-style feedback in natural language, providing users with personalized movement corrections. Authentication is handled using Auth0, enabling secure, user-specific sessions and feedback tracking across devices.

Challenges we ran into

Implementing real-time pose detection with sufficient accuracy was challenging, especially when dealing with different lighting conditions and camera angles. Optimizing the application for performance was crucial as processing video frames and running inference simultaneously is computationally intensive. Creating an algorithm that could provide meaningful feedback for different exercise types also required significant biomechanical research and testing.

Accomplishments that we're proud of

We successfully built a working prototype that provides accurate, real-time feedback on exercise form. The 3D visualization of body keypoints provides an intuitive representation of the user's movements, and our integration with the Groq API generates natural, helpful guidance that users can easily understand and implement. The system works across multiple exercise types and adapts to different body types.

What we learned

We gained critical insights into implementing machine learning models for real-time applications, balancing computational efficiency with accuracy in pose estimation. Learning to translate complex biomechanical data into actionable, user-friendly feedback proved challenging but rewarding. The project enhanced our understanding of creating intuitive healthcare interfaces while navigating the technical complexities of AI-driven movement analysis in diverse user environments.

What's next for Movemend

We plan to expand MoveMend to address more physical therapy protocol for different muscles ( injuries ) and provide more detailed analytics on users' progress over time. We want to implement a personalized feedback of physiotherapy based on users' form and progress. We're also exploring the possibility of creating a mobile application for greater accessibility and developing custom ML models specifically trained for physiotherapy applications.

Built With

• auth0
• blazepose
• groq
• javascript
• npm
• react
• react.js
• tensorflow.js
• webgl