Intelligent Automated Governance in E-Commerce Marketplace Catalogs: A Multi- Paradigm Asynchronous Architecture and RAG-Driven Decision Support System

Authors:

Kumar Sauryan,

Department of Computer Science & Engineering, Amity School of Engineering & Technology, Amity University Haryana, Gurgaon, India

Dr. Paras Chawla, Director and Professor, ASET (CSE Dept.), Amity University Haryana, Gurgaon, India

Dr. Dheeraj Singh, Professor, ASET, Amity University Haryana, Gurgaon, India

Abstract

Modern e-commerce ecosystems rely on rigid, automated, and algorithmic compliance protocols enforced at the product catalog layer via unique identifiers, such as Amazon Standard Identification Numbers (ASINs). Deviations from marketplace presentational and operational guidelines such as non-compliant title configurations or improper product canvas coverage— trigger instantaneous algorithmic suppression or listing deactivation. This immediately breaks transaction pipelines and induces severe financial volatility.

To mitigate these systemic operational risks, this paper presents SellerPro, a robust, multi- paradigm, AI-powered e-commerce account health tracking and risk prediction system. The architecture couples a high-throughput, non-blocking asynchronous Node.js and Express processing backend with a specialized Python-driven Retrieval-Augmented Generation (RAG) and Large Language Model (LLM) inference pipeline. By mapping unstructured catalog features into localized FAISS vector spaces, the system calculates cosine similarities against dense marketplace regulatory texts to extract contextually precise style guides. The system triages individual ASIN status profiles into three risk-classification tiers (Healthy, At Risk, or Unhealthy) and synthesizes deterministic, step-by-step remediation strategies.

Experimental validation demonstrates an optimized end-to-end latency of under 2.5 seconds and minimal memory footprint under high-concurrency catalog data ingestion.