Abstract: The rapid growth of digital fashion retail platforms has created a growing demand for recommendation engines capable of addressing the unique cultural, regional, and aesthetic diversity of Indian traditional attire. Existing systems predominantly rely on Western clothing datasets and fail to capture the nuanced interplay between Indian skin tones, body morphologies, regional textile traditions, and occasion-specific norms. This paper presents a comprehensive AI-based multimodal fashion recommendation system specifically designed for Indian traditional and regional clothing. The proposed architecture integrates a multimodal computer vision API to extract key biometric attributes — specifically skin tone classification from face photographs and body morphology estimation from full-body images — into a five-class skin tone schema and a seven-category body shape taxonomy. A custom rule-based content-based filtering engine, conditioned on an eight-dimensional user profile vector, then maps extracted biometric features to regionally appropriate Indian outfits across four geographic zones and eight occasion categories. An AI-assisted natural language generation module further enriches outputs with culturally contextualised descriptions, styling guidance, and personalised notes. The system is deployed as a full-stack web application using a React (Vite) frontend and a Python Flask backend communicating over a REST API. Experimental evaluation confirms sub-150 millisecond offline recommendation latency, strict cultural accuracy enforced by regional boundary filters, and positive user acceptance in preliminary testing. Comparative observations confirm that the decoupled biometric-extraction and recommendation architecture substantially reduces cultural misclassification over general-purpose generative approaches.

Keywords: Indian Fashion Recommendation; Computer Vision API; Content-Based Filtering; Skin Tone Detection; Body Morphology Classification; Regional Outfit Database; Flask; React; Multimodal AI; Cultural Recommendation System.

Downloads: | DOI: 10.17148/IARJSET.2026.133101