In recent years, yoga has emerged as a powerful tool for promoting holistic well-being. However, choosing appropriate yoga asanas tailored to an individual’s age, gender, and specific health concerns often requires expert consultation, which may not always be accessible. This paper presents a Smart Yoga Asana Recommender System that uses Natural Language Processing (NLP) and machine learning to generate personalized yoga recommendations. The system collects user inputs - age, gender, and a brief health issue description - and processes the textual data using a SentenceTransformer model to generate semantic embeddings. These embeddings are matched with a curated database of yoga asana descriptions using Facebook AI Similarity Search (FAISS), enabling fast and contextually relevant retrieval. Recommended asanas are compiled into a PDF report and automatically sent to the user’s email for convenient access. Built using Python, Flask, and MongoDB, the system offers a seamless web interface and integrates quote-based motivation for enhanced user engagement. This project combines traditional yogic wisdom with modern AI techniques to create a scalable and intelligent wellness tool, making expert-level yoga guidance accessible to individuals anytime, anywhere.
Introduction
In response to the growing importance of mental and physical wellness, this project introduces a Smart Yoga Asana Recommender System that provides personalized yoga pose recommendations based on a user’s age, gender, and health concerns. The system addresses the lack of expert yoga guidance, especially in remote areas, by leveraging AI, Natural Language Processing (NLP), and Machine Learning.
Key Components and Functionality:
User Input: Collected via a web form (age, gender, health concern, email).
NLP & Semantic Matching: A pre-trained SentenceTransformer (all-MiniLM-L6-v2) converts text input into vectors. These are matched with yoga pose descriptions using FAISS.
Database: A curated MongoDB dataset stores yoga poses, instructions, contraindications, demographic suitability, and Cloudinary-hosted images.
Filtering: Results are filtered based on age and gender.
Output: Recommended asanas are compiled into a visually rich PDF report using ReportLab and sent via email.
Architecture & Tools:
Backend: Python, Flask
Database: MongoDB
NLP: SentenceTransformer
Vector Matching: FAISS
PDF Generation: ReportLab
Email Delivery: SMTP
Visuals: Cloudinary
Literature Survey:
Previous studies explored yoga recommendation systems and mental health detection using AI, but most lacked personalization, demographic filtering, or automation. This project addresses these gaps with a fully automated, end-to-end recommendation system.
Methodology & Results:
The system was tested with different user profiles, showing its ability to adapt recommendations based on varying inputs. It generated and emailed relevant yoga pose PDFs with clear instructions and images, demonstrating accuracy, usability, and accessibility.
Key Features:
Semantic text understanding
Fast vector similarity search
Personalized and safe recommendations
Automated PDF creation and delivery
Motivational quotes and user-friendly UI
Conclusion
This research presents the development and implementation of a Smart Yoga Asana Recommender System that leverages NLP, vector similarity search, and modern web technologies to provide users with personalized yoga pose recommendations. By accepting natural language input and combining it with demographic filters, the system intelligently identifies relevant asanas from a curated database, compiles the results into a professionally formatted PDF, and delivers it via email — all within a matter of seconds.
The integration of tools such as SentenceTransformer, FAISS, MongoDB, and Flask allows the system to operate efficiently and scalably. The inclusion of visual content and motivational elements further enhances user engagement.
Overall, this work demonstrates how traditional wellness practices like yoga can be enhanced and scaled through artificial intelligence. It opens up possibilities for building accessible, data-driven health solutions that are both meaningful and easy to use.
References
[1] Deb, R. Khakharia, S. Shetty, A. Jadhav, and R. Sethi, \"Yoga pose recommender system for physical and mental illnesses,\" in IEEE Quantum Computing and Emerging Business Technology, ISBN: 979-8-3503-8427-7, 2024.
[2] S. Ramamoorthy, D. Singh, and S. Kar, \"Yoga recommendation system for the mental well-being of students using machine learning,\" in International Conference on Electronics, Communication and Aerospace Technology, ISBN: 978-1-6654-8271-4, 2023.
[3] A. Tom, G. Das, J. B. Hemmige, P. Deshpande, S. R. Bangshi, A. R. Paduri, and N. Darapaneni, \"Detecting mental health disorders using NLP,\" in IEEE Pune Section International Conference, ISBN: 979-8-3503-2420-4, 2024.
[4] V. Arya, N. Makattil, V. Sasikumar, V. Anuparvathi, and S. Khandare, \"Know your posture: Real-time posture detection and correction with yoga and exercise recommendations,\" in International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication, ISBN: 979-8-3503-1212-6, 2023.
[5] Nikitha, P., Nishanthini, S., Vinod, S., Mouleesh, Y., Silviya, S. H. A., & Sriman, B. “Intelligent Document Interaction with Advanced Vector Embeddings and FAISS-CPU Indexing,” Proceedings of the 2024 8th International Conference on Electronics, Communication and Aerospace Technology (ICECA), IEEE, Nov. 6–8, 2024. DOI: 10.1109/ICECA63461.2024.10800948.
[6] Kaur, Jagdish, and Kamboj, C. P. “Punjabi Grammar Checker: Automatic Correction of Punjabi Language Sentences Using Transformer,” Proceedings of the 2024 15th International Conference on Computing Communication and Networking Technologies (ICCCNT), IEEE, June 24–28, 2024. DOI: 10.1109/ICCCNT61001.2024.10725698.