Prakṛti Assessment Through Artificial Intelligence: Classical Foundations, Computational Approaches, And Future Directions In Personalised Āyurvedic Medicine — A Comprehensive Narrative Review

Abstract

Background- Prakṛti (psychophysiological constitution) is a foundational concept in Āyurveda that determines individual susceptibility to disease, drug response, and therapeutic outcomes. Traditional Prakṛti assessment relies on subjective clinical evaluation — physical examination, questionnaires, and pulse diagnosis — which suffers from inter-practitioner variability, lack of standardisation, and limited scalability. The rapid advancement of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) technologies offers transformative potential for objectifying, standardising, and scaling Prakṛti assessment through computational approaches. Objective- To comprehensively review the classical conceptualisation of Prakṛti in Āyurveda, systematically evaluate contemporary AI and ML approaches for Prakṛti classification including questionnaire-based models, pulse waveform analysis, facial image recognition, tongue diagnosis, genomic data mining, natural language processing, and wearable sensor integration; and to critically assess the current evidence, methodological challenges, ethical considerations, and future directions for AI-driven Prakṛti assessment in personalised medicine. Methods- Classical Āyurvedic texts (Charaka Samhitā Vimāna Sthāna 8, Suśruta Samhitā Śārīra Sthāna 4, Aṣṭāṅgahṛdayam Śārīra Sthāna 3) were reviewed for the Prakṛti framework. Published peer-reviewed literature on AI/ML applications in Prakṛti assessment was searched on PubMed, IEEE Xplore, Scopus, Google Scholar, DHARA, and AYUSH Research Portal using terms 'Prakriti machine learning', 'Ayurveda artificial intelligence', 'constitution classification deep learning', 'pulse diagnosis AI', 'Ayurvedic phenotyping computational', 'Prakriti genomics AI', and related terms. Publications from 2010 to 2025 were included. Grey literature including conference proceedings and dissertations were also considered. Results- Multiple AI/ML modalities have been applied to Prakṛti classification. Questionnaire-based models using Random Forest, SVM, and ANN achieve 75–93% classification accuracy for three-class (Vāta/Pitta/Kapha) discrimination. Pulse waveform analysis using piezoelectric/PPG sensors combined with CNN and LSTM architectures achieves 70–92% accuracy. Facial image analysis using computer vision and transfer learning (VGG16, ResNet) demonstrates 65–85% concordance with expert assessment. Genomic data integration using supervised learning on SNP profiles and gene expression data achieves significant Prakṛti discrimination. NLP-based analysis of classical texts enables automated feature extraction. Wearable IoT platforms combining multi-modal data show promising preliminary results. Key challenges include small datasets, class imbalance, lack of gold-standard labelling, and limited external validation. Conclusion - AI-driven Prakṛti assessment represents a high-potential convergence of traditional knowledge and modern computational science. While current evidence demonstrates feasibility and promising accuracy across multiple modalities, the field requires larger datasets, standardised clinical labelling protocols, multi-modal data fusion approaches, external validation across diverse populations, and robust ethical frameworks for deployment. The integration of AI-based Prakṛti tools into clinical Āyurvedic practice, telemedicine, preventive health screening, and pharmacogenomics could fundamentally transform personalised medicine delivery.