DEVELOP A CLOTHING E-COMMERCE WEBSITE WITH INTEGRATED VIRTUAL TRY-ON AND A SALES-ASSISTANCE CHATBOT

Abstract

The project "Develop a Clothing E-Commerce Website with Integrated Virtual Try-On and a Sales-Assistance Chatbot" addresses the need to optimize operations and automate customer care. The system uses a React (Next.js) frontend and a Node.js/Express backend secured with JWT authentication, with PostgreSQL (extended with pgvector) handling both business data and vector embeddings in a single database. Its core innovation is an AI shopping assistant built on the Gemini API using a RAG architecture: hybrid vector-and-keyword search grounds every product recommendation in real database catalog data, eliminating hallucinated prices or stock. A Function/Tool Calling mechanism lets the assistant search products by price, color, and size, add items to cart, and prepare checkout with the customer's chosen address, payment method, and voucher — automating the sales flow in real time. To keep the seemless experience of customers, the app implement model routing for auto-assign tasks to separate model based on task complexity and auto fallback to light model if usage limit reached. The system also complemented by a virtual try-on feature (photo-based garment compositing via a queued AI inference service and S3 storage) and integrated GHN shipping and VNPay/COD payment processing.Results confirm the system operates stably and securely, demonstrating the practicality of Generative AI — combined with retrieval grounding and tool use — in real-world e-commerce automation.