AI Fitness Trainer

• 

  Landing page

• 

  Workout with ai

• 

  Goals

• 

  Progress Tracking

Inspiration

The inspiration for AI Fitness Trainer came from our personal experience of working out without proper guidance. While many fitness apps and online videos exist, they lack real-time feedback and personalized correction, which often leads to incorrect posture and injuries. As AI & Data Science students, we wanted to explore how computer vision could be used to solve this real-world problem by creating an AI-powered trainer that can watch, analyze, and guide users during workouts using just a camera.

What it does

AI Fitness Trainer is a real-time fitness coaching web application that uses AI to analyze body posture, automatically count exercise repetitions, and provide instant voice feedback during workouts. Users can select a workout, turn on their camera, and start exercising while the system tracks reps, workout duration, and movement quality. The platform also supports goal setting, workout streaks, and progress tracking through a detailed history section.

How we built it

We built AI Fitness Trainer using a modern full-stack and AI-focused tech stack. The frontend is developed with React, TypeScript, and Vite, styled using Tailwind CSS and shadcn-ui for a clean and responsive interface. Supabase is used for authentication, user management, and storing workout data. For AI functionality, we integrated TensorFlow’s MoveNet pose-detection model to extract body keypoints from the live camera feed. These keypoints are processed in real time to analyze posture, count repetitions, and trigger voice feedback based on form deviations.

Challenges we ran into

One of the biggest challenges was handling noisy pose-detection data caused by lighting conditions, camera angles, and fast movements. We addressed this by applying temporal smoothing and threshold-based logic. Ensuring real-time performance while running AI inference in the browser was another challenge, as latency directly affects user experience. Designing accurate rep-counting logic and deciding when to trigger voice feedback without overwhelming the user also required careful tuning.

Accomplishments that we're proud of

We are proud of building a fully functional, end-to-end AI fitness application within a limited timeframe. Successfully running real-time pose estimation in the browser, delivering live voice feedback, and maintaining workout history without manual input were major milestones. Deploying a working beta version that users can access online was a significant achievement for our two-member team.

What we learned

Through this project, we gained hands-on experience with computer vision, pose estimation, and real-time AI systems. We learned how to integrate machine learning models into web applications, optimize performance, and manage noisy data. On the development side, we improved our skills in full-stack development, UI design, backend integration with Supabase, and team collaboration under hackathon pressure.

What's next for AI Fitness Trainer

In the future, we plan to add support for more exercises, personalized workout recommendations, and adaptive difficulty based on user performance. We also aim to improve form analysis accuracy, introduce advanced analytics, and enhance the feedback system using more intelligent AI models. Ultimately, we want AI Fitness Trainer to evolve into a complete, accessible personal training platform for everyone.

Built With

• react
• shadcn-ui
• supabase
• tailwind
• tensorflow.js
• vercel