Inspiration
One day, I was casually scrolling through reels when I came across a video that genuinely shook me. A person was performing a leg press, and suddenly his knee joints broke due to incorrect posture. It was traumatizing to watch. What stayed with me was the thought that this injury could have been avoided if someone had simply told him his posture was wrong — that he shouldn’t completely lock his legs, and that extending them to nearly 180 degrees was dangerous.
That moment made me stop and think. If there had been someone watching him and correcting his form in real time, this wouldn’t have happened. But the reality is, not everyone has access to a personal trainer. Most people work out alone, following online videos or guessing their form, because they can’t afford a gym where a trainer is constantly monitoring them.
This realization led to the idea of REPRIGHT — an AI-powered system that can act like a virtual trainer, watching every rep, identifying bad posture instantly, and giving real-time corrective feedback. The goal is simple: to help people work out safely, avoid preventable injuries, and make professional-level guidance accessible to everyone.
What it does
REPRIGHT is a web-based AI application that monitors exercise posture in real time using the user’s camera. It detects body landmarks through MediaPipe, calculates critical joint angles, and continuously evaluates whether the posture is correct or incorrect for a given exercise.
REPRIGHT:
- Highlights posture quality (good / bad) in real time_
- Provides live corrective feedback like “bend your knees more” or “keep your back straight”_
- Tracks key body landmarks and joint angles visually_
- Automatically counts reps, detecting upward and downward movement_
- Updates rep count instantly once a full repetition is completed_
- Adapts feedback dynamically based on the exercise being performed_
- In short, REPRIGHT acts as a virtual personal trainer that watches every rep and corrects form instantly._
How it was built
We built REPRIGHT using a combination of computer vision, AI/ML, and real-time web technologies, with a strong focus on low latency and continuous feedback.
The frontend accesses the user’s camera and streams video frames to the backend using WebSockets, enabling a persistent, two-way connection. Each frame is sent in real time, processed on the server, and the annotated frames and posture feedback are streamed back instantly to the frontend.
On the backend:
- MediaPipe is used for real-time body landmark detection_
- OpenCV processes each frame and overlays landmarks, angles, and posture indicators_
- AI/ML logic calculates critical joint angles and evaluates posture correctness_
- Rule-based validation is applied for each exercise to generate specific feedback_
- Movement direction (up/down) is tracked to accurately count repetitions_
On the frontend:
- The processed frames are rendered live_
- Rep count, posture status, joint angles, and corrective comments are displayed in real time_
This continuous WebSocket-based loop allows REPRIGHT to respond instantly to posture changes, ensuring minimal latency, smooth feedback, and a truly real-time virtual coaching experience.
Challenges we ran into
- Accurately detecting posture across different body types and camera angles
- Fine-tuning angle thresholds so feedback is helpful but not overly strict
- Avoiding false rep counts due to partial movements or shaky input
- Maintaining smooth real-time performance while processing video frames
- Designing feedback that is clear, actionable, and exercise-specific
Accomplishments that we're proud of
- managed to achieve real-time posture detection and correction using just a regular webcam, without any special hardware.
- We successfully built automatic rep counting by understanding when the user is moving up or down and detecting when a full repetition is completed.
- Instead of using generic posture rules, we designed exercise-specific feedback, so the guidance actually makes sense for each movement.
- We clearly visualized body landmarks and joint angles, helping users understand why their posture is right or wrong, not just whether it is.
What we learned
How small variations in joint angles can drastically affect exercise correctness
What's next for REPRIGHT
- Add support for more exercises and workout routines
- Introduce user profiles and progress tracking
- Provide personalised feedback depending upon person performing,his/her body shape,current physical state(** as it would not be good to discourage the person who has just started doing exercise,just because his posture was not perfect,as it is expected from a new beginner**)
Log in or sign up for Devpost to join the conversation.