AI GYM
PROJECT REPORT
Submitted in partial fulfillment of the requirements for the award of
Bachelor of Engineering
in
Computer Science
Visvesvaraya Technological University
Belagavi, Karnataka, 590 014
Alyster Benedict 2KE22CS187
Vijay Joshi 2KE22CS191
Tanay Kini 2KE22CS164
Ajit Adin 2KE22CS186
Submitted By
Dr. Geeta R. Bharamagoudar
professor
Department of Computer Science and Engineering
(NBA Accredited)
K. L. E. SOCIETY’S
K. L. E. INSTITUTE OF
TECHNOLOGY,
HUBBALLI
2024-2025
II
K. L. E. SOCIETY’S(9)
K. L. E. INSTITUTE OF TECHNOLOGY
Department of Name of the department
Engineering
CERTIFICATE
Certified that the project work entitled “AI GYM” is a bonafide work carried
out by Alyster Benedict (USN 2KE22CS187), Vijay Joshi (USN 2KE22Cs191),
Tanay Kini (USN 2KE22CS164), and Ajit Adin (USN 2KE22CS186), in partial
fulfilment for the award of degree of Bachelor of Engineering in VI Semester,
Computer Science and Engineering of Visvesvaraya Technological University,
Belagavi, during the year 2024-25. It is certified that all corrections/suggestions
indicated for internal assessment have been incorporated in the report deposited in
the department library. The project report has been approved as it satisfies the
academic requirements in respect of project work prescribed for the said degree.(12)
Dr. Geeta R. Dr. Rajesh Yakkundimath Dr. Manu T.M
Bharamagoudar HOD Principal
Name of the Examiners Signature with Date
1.
2.
III
DECLARATION
We, Alyster Benedict (USN 2KE22CS187), Vijay Joshi (USN 2KE22CS191)
Tanay Kini (USN 2KE22CS164) Ajit Adin (USN 2KE22CS186), students of VI
Semester B.E., K.L.E. Institute of Technology, Hubballi, hereby declare that the
project work has been carried out by us and submitted in partial fulfillment of the
requirements for the VI Semester degree of Bachelor of Engineering in Computer
Science of Visvesvaraya Technological University, Belagavi during academic year
2024-2025.
Date:06/05/2025
Place: Hubballi
Alyster Benedict
Vijay Joshi
Tanay Kini
Ajit Adin
IV
ACKNOWLEDGEMENT
We express our sincere gratitude to Dr. Geeta R. Bharamagoudar, our internal
guide, for her constant support, expert guidance, valuable suggestions, and
encouragement throughout the course of this project. Her feedback and direction have
been instrumental in helping us complete this work successfully.
We would also like to extend our heartfelt thanks to Dr. Rajesh Yakkundimath,
Head of the Department, Computer Science and Engineering, K.L.E. Institute of
Technology, for providing us with the opportunity and resources to undertake this
project, and for his continuous encouragement.
Our deepest appreciation goes to Dr. Manu T.M, Principal, K.L.E. Institute of
Technology, for creating an academic environment that fosters innovation and
learning, and for his support throughout our academic journey.
We are equally grateful to Dr. Dr. Yerriswamy T, Dean Academics, for his/her
valuable guidance, administrative support, and for facilitating the smooth progress of
our project work.
We would also like to thank all the faculty members of the Department of Computer
Science and Engineering for their suggestions, encouragement, and support at various
stages of the project. Special thanks to [Insert Name if you received support from any
external faculty or organization], for their insightful inputs and technical advice
during key phases of the project development.
Finally, we acknowledge the cooperation and teamwork of our group members—
Alyster Benedict (2KE22CS187), Vijay Joshi (2KE22CS191), Tanay Kini
(2KE22CS164), and Ajit Adin (2KE22CS186)—whose joint efforts made this project
possible.
V
K. L. E. SOCIETY’S
K. L. E. INSTITUTE OF TECHNOLOGY
ABSTRACT
AI-Fit is an intelligent fitness application designed to enhance personal training
through the integration of advanced computer vision, artificial intelligence, and real-
time interactive feedback. The primary goal of this project is to ensure that users
perform exercises with correct form by detecting posture deviations in real time and
providing immediate audio guidance. In doing so, the app aims to reduce the risk of
injury and improve the overall effectiveness of home workouts. Unlike traditional
fitness applications that focus primarily on calorie counting or static workout routines,
AI-Fit introduces a dynamic, data-driven, and interactive experience tailored to each
user’s physical attributes and fitness goals.
The background of this project is rooted in the increasing reliance on digital fitness
solutions, especially in post-pandemic times when many users have shifted to remote
or home-based workouts. However, current solutions often fall short in areas such as
personalized feedback, proper exercise execution, and holistic integration of support
features. Popular tools like Quick Pose and Neck Fit address isolated needs like pose
estimation or specific posture correction but lack comprehensive support systems. AI-
Fit was developed to address this gap by offering a cohesive platform that merges
real-time posture analysis, an AI-powered fitness chatbot, and a smart workout
recommendation engine into a single mobile solution.
The development process involved building a modular system using state-of-the-art
tools and algorithms. Media Pipe Blaze Pose was used for skeletal pose estimation,
allowing the app to track a user’s body movements and compare them with ideal
exercise postures. An algorithm evaluates these movements and delivers instant voice
feedback if incorrect form is detected. A virtual fitness assistant chatbot, implemented
using natural language processing techniques, allows users to ask questions, receive
encouragement, and get general workout advice. Additionally, the system includes a
workout recommendation engine that analyses user-specific data—such as age,
height, weight, and body photos—to generate a tailored monthly fitness plan. All
modules are supported by a robust backend that manages user profiles, tracks
progress, and stores relevant data securely.
Several test scenarios were designed to validate the effectiveness and usability of the
application. These include posture accuracy assessments using pre-labelled datasets,
chatbot interaction tests with varied user inputs, recommendation system validation
through user trials, latency checks for real-time feedback, and performance testing
across different Android devices.
VI
CONTENTS
Declaration
Certificate
Acknowledgement
Abstract
List of Figures
List of Tables
Chapter No. Name Page No.
1 INTRODUCTION 1
1.1 ……………… 1
1.2 ……………… 3
1.3 ……………… 4
1.4 ……………… 5
1.4.1 ……………… 5
1.4.2 ……………… 6
1.4.3 ……………… 6
1.4.4 ……………… 6
1.4.5 ……………… 6
1.5 ……………… 6
1.5.1 ……………… 7
1.5.2 ……………… 7
1.5.3 ……………… 8
1.6 Literature Survey 8
1.7 Motivation and Problem Definition 10
1.8 Objectives fulfilled 10
1.9 Scope and limitations 11
1.10 Relevance and Type of the project
1.11 Organization of the report 11
2 METHODOLOGY 12
2.1 ……………… 12
2.2 ……………… 13
2.3 ……………… 15
2.4 ……………… 15
2.4.1 ……………… 15
2.4.2……………… 16
2.4.3 ……………… 16
2.4.4 ……………… 19
2.4.5 ……………… 20
2.4.6 ……………… 20
2.4.7 ……………… 20
VII
1.1 Introduction
In recent years, the convergence of artificial intelligence (AI), computer vision, and
mobile computing has revolutionized several industries, with the fitness and wellness
sector being a key area of innovation. The increasing demand for home-based,
personalized, and interactive fitness solutions has accelerated the development of
intelligent systems capable of enhancing workout quality, ensuring safety, and
improving user motivation. Traditional fitness applications often provide generic
workout plans or simple tracking features such as calories burned and exercise
duration. However, these tools frequently lack the ability to evaluate user
performance in real time, particularly in terms of exercise posture and form—two
factors critical to preventing injury and maximizing the effectiveness of training
sessions.
The AI-Fit project addresses this gap by introducing a comprehensive fitness
application that leverages real-time posture detection, AI-driven audio feedback, and
an intelligent virtual assistant to create a smart and interactive workout environment.
The system’s computer vision module continuously monitors the user's body position
using skeletal pose estimation algorithms, immediately identifying and correcting
form deviations through spoken cues. This feedback loop enhances user safety and
performance by promoting optimal body alignment during each exercise.
To further increase engagement and personalization, AI-Fit incorporates a chatbot
assistant capable of answering fitness-related queries, offering motivational prompts,
and guiding users through their routines. The chatbot enhances interactivity and
serves as a virtual fitness companion, simulating a personal trainer experience without
the associated costs. Additionally, the application features an advanced workout
recommendation system that utilizes user-provided data—such as height, weight, age,
and body photographs—to generate tailored monthly workout plans. This ensures that
each routine is specifically designed to match the user’s current fitness level and
goals, enabling gradual and safe progress over time.
The architecture of AI-Fit is modular, combining distinct yet interconnected
components for posture detection, chatbot communication, mobile interface
development, and backend data management. This design supports scalability,
maintainability, and future integration of additional features. By uniting these
technologies into one accessible platform, the project demonstrates how
interdisciplinary innovation can enhance personal fitness experiences and promote
healthier lifestyles.
Ultimately, the AI-Fit application aims to set a new benchmark in digital fitness by
delivering a solution that is not only intelligent and interactive but also personalized,
cost-effective, and user-friendly. This introduction lays the foundation for exploring
VIII
the design, methodology, and results that validate the potential of AI-Fit to reshape
the digital fitness landscape.
1.2 Literature Survey
The integration of artificial intelligence and computer vision into fitness and wellness
applications has grown rapidly over the last few years, driven by increasing demand
for personalized and home-based exercise solutions. A substantial body of research
highlights the importance of correct posture and form during physical activity, with
studies confirming that improper movements can significantly reduce the
effectiveness of workouts and increase the risk of injury. As a result, the use of pose
estimation algorithms and machine learning models in digital fitness tools has
emerged as a promising area for innovation.
One of the major advancements in this domain is the development of real-time human
pose estimation frameworks such as Media Pipe Blaze Pose, Open Pose, and Pose
Net. These technologies enable accurate skeletal tracking using a device’s camera and
are commonly employed in exercise monitoring systems. Blaze Pose, for example,
offers high-performance real-time tracking suitable for mobile applications and has
become a popular choice for developers aiming to implement fitness form correction
systems. Studies involving pose estimation systems demonstrate how machine
learning can classify and assess user posture by comparing detected body landmarks
to predefined ideal poses for specific exercises.
Commercially, tools like Quick Pose have leveraged these technologies to build
Software Development Kits (SDKs) that offer plug-and-play pose estimation
capabilities for fitness app developers. Quick Pose combines Media Pipe with Blaze
Pose to provide reliable real-time skeletal tracking and posture correction. It also
offers features like exercise counters, feedback systems, and pose validation APIs.
While Quick Pose is technically sound and easy to integrate, it focuses primarily on
posture detection and lacks broader support for personalized workout planning or
intelligent assistant integration.
Another relevant commercial example is Neck Fit AI Workout, which is tailored to
improve neck posture and correct forward head positioning. While effective in its
niche, this application does not offer holistic fitness guidance or adapt workout plans
based on diverse user metrics. A significant portion of current fitness apps either
focus on posture correction or general activity tracking (e.g., step counters, calorie
calculators), without integrating these features into a personalized and interactive
workout environment.
The reviewed literature and existing applications reveal clear limitations in the current
market. Most notably, there is an absence of fully integrated solutions that offer
personalized workout plans, real-time posture correction, and interactive AI
chatbot support in one unified platform. Furthermore, the high cost of certain
IX
advanced fitness apps and the lack of modular adaptability also create accessibility
barriers for many users. While some research prototypes address individual technical
challenges, few have progressed into complete, user-friendly applications that merge
all key components into one cohesive experience.
In summary, the literature indicates a strong foundation in pose estimation and AI-
based feedback systems, yet highlights significant opportunities for innovation in
personalized fitness applications. The AI-Fit project seeks to fill this gap by
combining intelligent posture analysis, an AI-powered chatbot assistant, and a
dynamic, user-specific workout recommendation system—all delivered through an
accessible and integrated Android application.
X
1.3 REFERENCES
1. P. Kumar and A. Kumar, Personalized Fitness Guidance Using AI-Driven
Recommendation System, Journal of Engineering and Science Publication, vol.
15, no. 6, 2024.
2. A. S. R. Research Group, Pose Estimation and Virtual Gym Assistant Using
Media Pipe and Machine Learning, ResearchGate, 2023.
3. AI Fitness Workout Assistant Using NLP Techniques, International Journal of
Scientific & Technology Research (IJSAT), Jan. 2025.
4. X. Zhang, Y. Chen, and L. Wu, A Review on Computer Vision Technology for
Physical Exercise Monitoring, Algorithms, vol. 15, no. 12, 2022.
5. Quick Pose, Pose Estimation/Detection SDK for iOS, QuickPose.ai, 2024.
[Online]. Available: https://quickpose.ai
6. E. Trigo, Posture: Neck Fit AI Workout, Google Play Store, 2024. [Online].
Available: https://play.google.com/store/apps/details?id=com.neckfit
7. M. Scholder, R. Heinz, and T. Becker, Personalized Physical Activity
Coaching: A Machine Learning Approach, ResearchGate, 2017.
8. S. Patel and R. Singh, AI-Driven Personalized Fitness Coaching with Body
Type-Based Recommendations, International Journal of Preventive Medicine
and Health, vol. 5, no. 1, 2025.
9. Y. Singh, Working on a Fitness Trainer Bot, Rasa Forum, Jan. 2021. [Online].
Available: https://forum.rasa.com
10. O. Hruby, chatbot-celestines, GitHub, 2023. [Online]. Available:
https://github.com/ochorup
11. D. Mohanty, Real-Time Pose Estimation for Fitness Applications with AI and
Computer Vision, Medium, Oct. 2024. [Online]. Available:
https://medium.com
12. Grand View Research, Fitness App Market Size & Share | Industry Report,
2030, Apr. 2024. [Online]. Available: https://www.grandviewresearch.com
13. J. Schiffer, AI Health Coaches Are Coming Soon to a Device Near You, TIME
Magazine, Nov. 2023.
14. A. Ray, I Used ChatGPT as My Personal Trainer. It Didn’t Go Well, TIME
Magazine, Apr. 2024.
15. NCBI, Using Artificial Intelligence for Exercise Prescription in Personalized
Programs, PubMed Central, 2024. [Online]. Available:
https://www.ncbi.nlm.nih.gov
16. K. Lee and J. Park, Real-Time Workout Posture Correction using OpenCV
and Media Pipe, KI-IT Conference Proceedings, 2024.
XI
1.4 Conclusion and Future Scope
The AI-Fit project successfully demonstrates the integration of advanced computer
vision, AI-driven chatbots, and intelligent recommendation systems into a cohesive
fitness application. By focusing on real-time posture detection and immediate
feedback, the application enhances workout accuracy and reduces the risk of injury.
Additionally, the use of a personalized workout planner based on user metrics such as
height, weight, age, and body image add a valuable layer of customization that most
existing solutions lack. The implementation of an AI chatbot for instant user
interaction further increases engagement and provides support similar to that of a
virtual trainer. Collectively, these components establish AI-Fit as a modern, holistic
fitness companion capable of adapting to a wide range of user needs.
Looking ahead, the future scope of this project includes incorporating wearable device
integration for even more accurate motion tracking and heart-rate-based training
feedback. Another promising direction is using deep learning models to better predict
user performance trends and dynamically adjust workout difficulty. Multilingual
support for the chatbot and adding voice-controlled functionality could further
improve accessibility. Additionally, cloud-based user profile storage and progress
tracking across devices would elevate the application's flexibility and long-term user
retention. With these enhancements, AI-Fit has the potential to evolve into a robust
digital fitness ecosystem that rivals premium subscription platforms, making
personalized training both accessible and effective.
XII
