HealthMate: An AI-Powered Real-Time Fitness Coaching System with Pose Estimation and Nutrition Planning

Authors: Khan Mohammed Ashaaf , Kartik Limbachiya, Suyash Katkam, Anugrah Kulkarni, Prof. Pallavi Khodke

DOI Link: https://doi.org/10.22214/ijraset.2026.80172

Abstract

The emergence of mobile health apps has resulted in an ecosystem of siloed products, where exercise training and nutritional advice are delivered independently without any interaction between the two systems. HealthMate seeks to overcome this issue through the implementation of real-time AI-powered posture analysis, repetition counting, and goal-oriented nutritional planning into one unified browser application. Using the MediaPipeBlazePose model in the frontend part, HealthMate detects 33 skeletal key points per video frame and sends normalized coordinate payloads to FastAPI backend through persistent WebSocket connections. The proposed repetition counting system relies on calculating the angles of elbows, shoulders, hips, and knees using the cosine rule pipeline and classifying repetitions according to empirically determined thresholds as either correct or not correct, providing the user with detailed corrective feedback in both cases. Firebase Firestore provides the secure storage and management of sessions and users\' information. During experimental evaluation involving 5 participants and 10 different sessions under various lighting conditions, the repetition counting showed 94.6% accuracy and mean end-to-end delay of 274 ms (95%ile – 298 ms), yielding a System Usability Scale of 78.3 (good).

Introduction

HealthMate is a browser-based health application that integrates AI-powered exercise tracking and personalized nutrition planning into a single platform, addressing the fragmentation of traditional fitness and nutrition apps. It uses the MediaPipe BlazePose model to detect 33 body keypoints in real time and transmits normalized skeletal data to a FastAPI backend via WebSocket connections. The system calculates joint angles using the cosine rule to accurately count exercise repetitions and assess posture, providing users with real-time corrective feedback. User data and workout sessions are securely stored using Firebase Firestore. Experimental testing with five participants across ten sessions under different lighting conditions demonstrated 94.6% repetition-counting accuracy, an average end-to-end latency of 274 ms (95th percentile: 298 ms), and a System Usability Scale (SUS) score of 78.3, indicating good usability and effectiveness.

Conclusion

In this study, a fitness coach application named HealthMate has been proposed to provide live feedback by using pose-estimation technology for exercising analysis and nutrition management based on the goals. HealthMate successfully achieved the accuracy of repetitions at 94.6% with a mean delay of feedback via WebSocket at 274 ms and a usability score on System Usability Scale of 78.3, satisfying all design criteria. The key technical innovation is the proof-of-concept that a cosine rule-based joint angle pipeline running on the server in Python, taking in 2-D landmarks payloads via WebSocket from a Web hosted MediaPipe pipeline, can provide biomechanically relevant exercise feedback under sub-300 ms latency with off-the-shelf hardware and no client installation. The following are some suggestions for future research. It will be necessary to enrich the exercise library by adding exercises such as deadlifts, lunges, and overhead press; implement dynamic nutrition planning based on parameters such as body mass index, calorie consumption, and physical activity per day; introduce an interactive chatbot that provides verbal assistance to the user through natural language; and design a dashboard that helps track progress in terms of form improvement over several weeks or training periods.

References

[1] Grand View Research, “Fitness App Market Size, Share & Trends Analysis Report,” San Francisco, CA, USA, Tech. Rep. GVR-3-68038-944-1, 2024. [2] L. Noteboom et al., “Factors Associated with Gym-Based Fitness Injuries: A Case-Control Study,” Int. J. Environ. Res. Public Health, vol. 20, no. 5, p. 4434, Feb. 2023. [3] V. Bazarevsky et al., “BlazePose: On-device Real-time Body Pose Tracking,” arXiv:2006.10204, Jun. 2020. [4] Z. Cao et al., “OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 1, pp. 172–186, Jan. 2021. [5] C. Lugaresi et al., “MediaPipe: A Framework for Building Perception Pipelines,” arXiv:1906.08172, Jun. 2019. [6] A. Eguchi et al., “The Efficacy of an mHealth App in Facilitating Weight Loss: A Randomized Controlled Study,” JMIR mHealth uHealth, vol. 11, e43530, Mar. 2023. [7] A. M. Saad et al., “Diet Engine: A Real-Time Food Nutrition Assistant,” in Proc. Int. Conf. Computer, Electrical and Communication Engineering (ICCECE), Kolkata, India, 2025. [8] T. Veneman et al., “A Mobile Health App for Home-Based Aerobic Exercise in Neuromuscular Disease,” J. NeuroEngineeringRehabil., vol. 21, no. 1, p. 28, Feb. 2024. [9] J. Liu and Y. Zhou, “Smart Fitness with YOLO-Fit IoT: Real-Time Pose Analysis and Personalized Training via IoT and Reinforcement Learning,” IEEE Internet Things J., vol. 12, no. 3, pp. 1102–1115, Feb. 2025. [10] C. M. Beattie et al., “Environmental and Personal Factors That Influence Gym Members with Disabilities,” Disabil. Rehabil., vol. 46, no. 11, pp. 2451–2460, 2024. [11] P. J. Marcos-Pardo et al., “Optimizing Outdoor Fitness Equipment Training for Older Adults,” Front. Physiol., vol. 14, 1181654, May 2023. [12] J. Nielsen, “Response Times: The 3 Important Limits,” Nielsen Norman Group, 1993. [13] S. Abeywickrama et al., “Real-Time Exercise Repetition Detection Using Smartphone Accelerometers and Pose Estimation,” in Proc. IEEE 17th Int. Conf. Industrial Informatics (INDIN), Helsinki, Finland, 2019, pp. 1562–1567. [14] A. Bangor, P. Kortum, and J. Miller, “Determining What Individual SUS Scores Mean: Adding an Adjective Rating Scale,” J. Usability Studies, vol. 4, no. 3, pp. 114–123, May 2009. [15] S. Fida et al., “Wearable-Based Posture Recognition in Construction Workers,” IEEE Access, vol. 8, pp. 150462–150474, 2020.

Copyright

Copyright © 2026 Khan Mohammed Ashaaf , Kartik Limbachiya, Suyash Katkam, Anugrah Kulkarni, Prof. Pallavi Khodke. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download Paper

Paper Id : IJRASET80172

Publish Date : 2026-04-14

ISSN : 2321-9653

Publisher Name : IJRASET

DOI Link : Click Here