RepRight | Devpost

What and Why

RepRight is an AI-assisted training tool designed to make strength training less intimidating for beginners. Many people, particularly women, beginners, and older generations, avoid gyms due to social pressure, fear of injury, or lack of confidence in their form.

Our software provides real-time feedback on exercise form, helping users learn correct technique safely and privately. By removing the fear of “doing it wrong”, Rep Right encourages more people to start strength training, build confidence, and improve their long-term health.

Theme: Flipping the Script

Strength training and gym culture are often associated with intimidating environments, unrealistic body expectations, and a perception that weightlifting is “not for everyone.” This can discourage many people, especially women, beginners, older adults, and other underrepresented groups, from getting started.

RepRight flips the script by making training accessible, supportive, and beginner-friendly. Instead of relying on expensive personal trainers, prior gym experience, or confidence in a male-dominated environment, our app empowers anyone to exercise safely with AI guidance and real-time feedback.

Through intentional design choices, such as a female training character, approachable interface, and clear visual cues, we challenge traditional stereotypes about who can participate in strength training. The app encourages users to focus on personal progress, confidence, and safety, rather than comparing themselves to gym norms or aesthetics.

By focusing on confidence, accessibility, injury prevention, and inclusivity, Rep Right actively subverts societal expectations and the conventional gym narrative. It reframes strength training as a space for all bodies, abilities, and experience levels, creating a fitness experience that is welcoming, empowering, and equitable.

RepRight does more than just teach exercises; it redefines what a gym culture can be. By flipping the script on intimidation and exclusion, it inspires more people to confidently take their first step towards a healthier lifestyle.

Most Technically Impressive

RepRight integrates computer vision pose estimation with inertial sensor data to analyse exercise form in real time. We used OpenPose for body pose estimation, running a pre-trained deep learning model to detect 2D skeletal keypoints (major joints such as shoulders, elbows, hips, and knees) from camera input. These keypoints allow us to reconstruct the user’s body posture and analyse joint angles during exercises.

The system was implemented in MATLAB, where we processed the OpenPose output to calculate joint angle trajectories and movement patterns. By comparing these values against predefined biomechanical thresholds for each exercise, the system can detect incorrect form such as poor alignment or incomplete range of motion.

To enhance motion tracking, we also incorporated IMU sensor data from smartphones using MATLAB Mobile, allowing us to capture accelerometer and gyroscope data during exercises. This additional data provides insight into movement, improving the robustness of movement detection and injury prevention.

The application interface was developed using MATLAB App Designer, which integrates the computer vision pipeline, sensor data stream, and feedback system into a real-time interactive training tool.

Best User Interface

Our interface was designed to feel approachable, supportive, and unintimidating, particularly for beginners who may feel overwhelmed by traditional fitness environments or complex training apps. Each exercise mode includes clear instructions, simple visual guidance, and short descriptions explaining how to perform movements safely and which muscle groups are being targeted. We intentionally kept text concise and readable so users can quickly understand instructions without disrupting their workout.

We also made deliberate visual design choices to make the interface feel more inclusive. The on-screen training character is represented as a woman, helping to challenge the common perception that strength training tools are designed primarily for men. This aligns with our goal of encouraging more women and underrepresented groups to feel comfortable engaging with strength training. Our colour palette centres around purple. We chose this to create a visual identity that feels welcoming and supportive, rather than aggressive or overly performance-focused like many traditional fitness apps.

The interface itself was developed using MATLAB App Designer, allowing us to build a clean layout that integrates live camera feedback, pose visualisation, and form guidance in one place. Real-time feedback is displayed clearly so users can easily understand when they are performing an exercise correctly or when adjustments are needed. Different exercise intensity levels allows users to choose between beginner, intermediate, and advanced variations. This helps users gradually progress their training while maintaining correct form and reducing the risk of injury. The clean and minimal layout ensures the application is easy to navigate and avoids overwhelming the user. By prioritising simple visual elements, clear spacing, and intuitive controls, the interface allows users to focus on their exercise and real-time feedback rather than struggling to understand the app.

Overall, every design decision was made intentionally to ensure the product feels accessible, inclusive, and easy to use, helping users focus on building confidence in their training rather than feeling intimidated by technology or gym culture.

Societal Impact

RepRight aims to reduce barriers to exercise and make strength training accessible and welcoming to a wider range of people. Many individuals avoid gyms due to fear of judgment, lack of knowledge about proper form, or concerns about injury. These barriers disproportionately affect beginners, women, older adults, and people from underrepresented groups, who may feel less confident in traditional fitness environments.

By providing real-time feedback on exercise form, RepRight allows users to learn and practise movements in a supportive, private environment. This helps build confidence and encourages users to develop safe, sustainable exercise habits without requiring access to expensive personal trainers.

Our project also promotes health education, helping users understand how to perform exercises correctly, which muscle groups are targeted, and how to progress safely. This encourages better long-term exercise habits and reduces the risk of injury caused by improper technique.

RepRight actively challenges traditional fitness culture. Through intentional design choices, such as representing a female training character, using a supportive colour palette, and maintaining a beginner-friendly interface, we demonstrate that strength training is for everyone. By normalising participation from women, older adults, and other underrepresented groups, Rep Right helps subvert societal expectations about who “belongs” in the gym.

We also aim to improve accessibility and inclusivity in fitness technology. Many current fitness tools are designed for experienced gym-goers, whereas Rep Right focuses on beginners, people with injuries, and those who may feel excluded from traditional fitness culture. Future developments, such as adaptive exercises for wheelchair users and personalised recommendations based on injuries, will further expand access.

The long-term social impact of RepRight includes:

Encouraging more people to engage in regular physical activity
Building confidence and reducing intimidation for beginners and underrepresented groups
Reducing exercise-related injuries through better form education
Supporting lifelong healthy habits by making safe exercise accessible
Promoting inclusivity and changing the cultural narrative around strength training

We could measure this impact through user engagement, improvements in exercise form over time, workout completion rates, user-reported confidence, and adoption among underrepresented groups. Over time, we hope RepRight will foster a more inclusive and supportive fitness culture, empowering people to exercise safely and confidently, regardless of their background or prior experience.

Most Sustainable

RepRight promotes sustainability by making fitness accessible using devices people already own, such as smartphones and webcams. Rather than requiring specialised gym equipment, wearables, or expensive hardware, our system uses a standard camera and smartphone sensors to analyse exercise form. This approach lowers the barrier to entry for users while reducing the need for additional manufactured devices.

Encouraging people to exercise at home or in accessible environments reduces the need for travel to gyms, contributing to lower energy consumption and a smaller carbon footprint.

We also optimised the AI model by stripping it back to only the components necessary for our application. By reducing the complexity of the pose estimation pipeline and limiting the number of keypoints and computations processed, we were able to significantly decrease the computational load. This reduces the processing power and energy required to run the model in real time, making the system more efficient and accessible on standard consumer hardware. By prioritising lightweight AI and efficient processing, our approach supports more sustainable computing practices while still delivering accurate real-time feedback for users.

In the future, we would love to scale RepRight using energy-efficient cloud infrastructure powered by renewable energy, allowing our AI models to run at scale while minimising environmental impact. By combining digital fitness tools with sustainable computing practices, we aim to create a solution that supports both personal health and environmental sustainability.

Best Use of AI

RepRight uses AI-powered pose estimation to act as a real-time virtual personal trainer. We implemented OpenPose, a deep neural network trained on large-scale human pose datasets, to detect human skeletal keypoints from live camera input. The model outputs the spatial coordinates of key joints across the body, allowing us to reconstruct a digital skeleton of the user.

Within MATLAB, these keypoints are processed to compute joint angles, limb alignment, and movement trajectories during exercises. AI-based pose detection allows the system to interpret complex body movements and determine whether an exercise is being performed with correct technique.

To improve movement understanding, the system also integrates IMU sensor data from MATLAB Mobile, capturing accelerometer and gyroscope readings from a smartphone attached to the user. This enables a simple form of sensor fusion, combining visual pose data with motion data to better detect movement quality. By translating this complex AI and sensor data into simple corrective feedback, RepRight enables users to safely learn proper exercise form without needing a human trainer.

Challenges We Faced

Developing RepRight involved several technical and design challenges, particularly because we wanted to build the entire system in MATLAB for easier integration of the front-end interface with the back-end AI processing.

One of the main challenges was maintaining a reasonable frame rate. OpenPose and the associated MATLAB toolboxes are computationally heavy, and processing live video frames while performing pose estimation in real time caused noticeable lag. We had to carefully optimise the code and reduce unnecessary computations to keep the app responsive for the user.

Another was accurately mapping the limbs and calculating joint angles. Ensuring that the angles matched the user’s real movement required careful calibration and testing, especially when combining OpenPose keypoints with IMU sensor data from MATLAB Mobile. Small errors in joint mapping could result in incorrect feedback, so a lot of effort went into verifying the skeleton overlay and joint angle calculations.

Building a clean, modular UI that could display live skeleton data, feedback prompts, and exercise instructions simultaneously was also challenging. MATLAB App Designer is powerful but can be unintuitive for dynamic, real-time visualisation. We had to manage multiple live plots, overlays, and user inputs while keeping the interface clear, responsive, and beginner-friendly.

Integrating OpenPose into MATLAB was another hurdle. OpenPose is primarily Python-based, so we faced compatibility issues, including loading the ONNX model correctly and converting outputs into MATLAB-readable formats. Ensuring the skeleton detection was accurate and synchronised with our IMU data required extensive testing and troubleshooting.

Finally, collaboration and code management presented challenges. Because the project involved multiple team members, we needed to maintain clean, well-documented, and modular code to allow everyone to work on different parts of the system (UI, AI processing, and sensor integration) without conflicts.

Despite these challenges, tackling them helped us develop robust, optimised code and an intuitive interface, and gave the team valuable experience integrating complex AI models, sensor data, and real-time visualisation into a single MATLAB application.

Future Development

While RepRight currently focuses on demonstrating real-time form feedback for a small set of exercises, there are many opportunities to expand the platform in the future.

One key area of development would be adding a wider range of exercises across different training categories such as strength training, mobility, and rehabilitation. We are also interested in making the platform more inclusive and accessible, particularly by expanding support for wheelchair users and adaptive exercises. By adapting our pose detection and motion analysis models, Rep Right could provide guidance for seated workouts and mobility-focused training, helping ensure that more people can benefit from safe and supported exercise.

OpenPose can also handle multiple people, so adding in multiplayer modes would be a fun addition to gamify exercise to increase enjoyment.

Another future feature would allow users to input existing injuries or physical limitations when setting up their profile. The system could then avoid recommending exercises that may aggravate those conditions and instead suggest safer alternatives. This would help create more personalised and injury-aware training plans.

We would also like to expand the AI-driven recommendation system so the app can suggest exercises and workouts tailored to the user’s goals, ability level, and progress over time. By analysing previous sessions and movement patterns, the system could provide increasingly personalised training guidance.

Finally, we plan to introduce progress tracking and performance reports, allowing users to monitor improvements in form, consistency, and workout completion. These reports could help users better understand their progress and stay motivated while building long-term healthy exercise habits.