Conversational Fashion Outfit Generator using AI/ML and Natural Language Processing

Abstract

While digital apparel marketplaces feature expansive item catalogues, they simultaneously induce choice overload for end-users. Standard query interfaces heavily depend on static taxonomy structures and rigid checkbox criteria, completely isolating the platform from natural, multi-attribute human shopping expressions. To mitigate this transactional friction, this study presents an automated, text-and-voice-driven conversational outfit selection engine. The system processes raw user intent using the spaCy linguistic pipeline to isolate relevant nouns, adjectives, and contextual modifiers. These extracted fashion attributes are converted into high-dimensional numerical coordinates using a Term Frequency-Inverse Document Frequency (TF-IDF) vector space model. Similarity calculations are executed via a Nearest Neighbours algorithm applying cosine metrics against an index of over 44,000 fashion products. Implemented using a responsive Stream lit interface, the system integrates immediate voice-to-text processing, strict gender subset constraints, and custom input validation error-trapping routines. Empirical testing confirms rapid recommendation generation and accurate token targeting, establishing an optimized design for lightweight, hands-free conversational commerce applications.