YogiSync

home and chatbot
home page features
practice section
learn pose
warmup
practice pose
playlist
report
AI Coach

Inspiration

The inspiration for YogiSync came from our own challenges with maintaining proper form during yoga practice. We noticed that many practitioners, especially beginners, struggle to self-correct without immediate feedback from a teacher. Combining our passion for technology with a love for yoga, we envisioned a tool that could offer personalized guidance anytime, anywhere—making yoga more accessible and effective.

What it does

YogiSync leverages cutting edge technologies to create an all purpose yoga guidance solution. It provides the user with the following features-

Detects and Classifies Poses: Uses MediaPipe for pose detection and a custom-trained ML model to classify various yoga poses. This allows users to self correct and learn new workouts
Provides Immediate Feedback: Overlays skeleton visualizations on the live video feed and offers corrective instructions to ensure proper alignment.
Chatbot that can recommend you ideal yoga poses based on your needs and goals and makes playlists for the user
AI yoga coach that generates a personalized yoga and dietary routine for the user ## How we built it
We began by collecting our own dataset of yoga poses, ensuring diversity in body types, lighting conditions, and angles. This custom dataset allowed us to train a machine learning model tailored specifically for yoga pose detection.
Using Python, we pre-processed the collected data and trained a custom ML model to classify various yoga poses. We used popular libraries (such as scikit-learn or TensorFlow/Keras) for model training and then serialized the model with pickle, making it easy to load and integrate into our Flask backend.
We leveraged Google’s MediaPipe library to extract pose landmarks from webcam video streams. MediaPipe provided us with accurate, real-time joint detection, which was crucial for assessing the user’s pose.
The extracted landmarks are fed into our custom ML model to classify the pose. We also implemented additional angle logic for real-time feedback on pose correctness, ensuring users receive actionable guidance.
Our backend is built with Flask, which serves as the core framework for handling HTTP requests, managing user sessions, and coordinating real-time video analysis.
To enhance the user experience, we integrated a conversational LLM. Using Google’s Gemini LLM, our chatbot functions as an AI yoga coach.
Our frontend is built using Bootstrap, ensuring a responsive and modern design that works across devices. We designed a clean and intuitive interface

Challenges we ran into

Achieving the right balance between speed and accuracy for live video analysis was challenging. Optimizing frame processing and ensuring consistent performance across varying lighting conditions required significant fine-tuning.
Merging computer vision, machine learning, and interactive frontend elements into one cohesive application presented several hurdles, particularly with synchronizing real-time feedback and maintaining a smooth user experience. ## Accomplishments that we're proud of
Successfully integrating MediaPipe and our ML model to deliver accurate, real-time feedback on yoga poses.
Building an application that spans from the backend (Flask, ML model integration) to a polished, responsive frontend that works seamlessly across devices.
Integrating GenAI features that actually help make people's lives better

What we learned

Advanced Computer Vision Techniques - Working with MediaPipe and custom ML models deepened our understanding of real-time image processing and pose estimation.
Full-Stack Integration - We learned how to bridge complex backend functionalities with a dynamic, user-friendly frontend, ensuring a seamless user experience.
Leveraging GenAI to develop uselful features

What's next for YogiSync

We plan to introduce more personalized workout and nutrition plans based on user progress and feedback, making the app even more tailored to individual needs.
Adding more yoga poses and integrating advanced posture analysis to provide comprehensive guidance for users of all levels.
Developing a social component that allows users to share their progress, join group challenges, and foster a supportive yoga community.
Incorporating data from wearables (such as heart rate and activity trackers) to provide more holistic, real-time feedback during yoga practice.

Built With

Submitted to

BrickHack 11

Created by

I designed the UI/UX for our website, creating intuitive and visually engaging interface with a strong focus on usability and user experience. Additionally, I contributed to front-end development, leading the integration of design components to ensure seamless functionality and visual coherence.

Tanvi Chandan
I worked on the frontend as well and backend. I worked on the development of LLM based yoga coach and chatbot. Additionally, I spearheaded the integration of design components in the frontend.

Anusha Seshadri
I spearheaded the backend development, integrating Mediapipe for real-time pose detection. This involved extracting and pre-processing data using Mediapipe, followed by training multiple machine learning models. The Random Forest model emerged as the top performer, achieving the lowest test loss. Additionally, I developed all necessary APIs to facilitate seamless integration with the front-end via Flask.

Shubh Sehgal
I worked on the backend. I integrated the custom-trained ML model for pose classification and implemented logic for corrective feedback. Additionally, I managed the MediaPipe pipeline for pose detection and ensured seamless backend-frontend communication.

Iyashi Pal