PERSONALIZED FASHION ASSISTANT
Mrs. BHAVITHIRA V AP/AI&DS
ABISHEK M DHARSHAN K J JERSIL G M KARTHICK B
BACHELOR OF TECHNOLOGY – DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND DATA SCIENCE (FINAL YEAR)
SRI SHAKTHI INSTITUTE OF ENGINEERING AND TECHNOLOGY (AUTONOMOUS) COIMBATORE – 641062
ABSTRACT :
FitGenie is an AI-powered personalized fashion assistant system designed to deliver intelligent, context-aware outfit recommendations tailored to individual users. The system integrates computer vision, machine learning, and large language models into a unified platform that functions as a smart digital wardrobe and virtual stylist. It collects user-specific attributes such as age, gender, skin tone, body type, and style preferences, and processes uploaded clothing images using a fine-tuned Vision Transformer deep learning model to extract garment category, color, pattern, and fabric attributes. The resulting feature vectors are indexed in a FAISS vector database for sub-millisecond similarity retrieval, while structured metadata is stored in PostgreSQL.
The core recommendation engine employs a Retrieval-Augmented Generation (RAG) pipeline that integrates real-time weather data, calendar event details, and current fashion trends to generate personalized outfit suggestions with natural language rationale through a Large Language Model. A virtual try-on module powered by a latent diffusion model enables photorealistic simulation of selected garments on user images through skeletal pose estimation and thin-plate spline warping. A continuous feedback mechanism updates a User Preference Matrix to improve recommendations over time, while additional features include outfit history tracking, closet gap analysis, and calendar-based scheduling.
KEYWORDS: Personalized Fashion Assistant, FitGenie, Computer Vision, Retrieval- Augmented Generation (RAG), FAISS Vector Database, Large Language Model (LLM), Virtual Try-On, Diffusion Model, Smart Wardrobe, Deep Learning, Context-Aware Recommendation, Generative AI, Fashion AI, PyTorch, FastAPI.
