This document describes a real-time product recommendation system called KijiShopping that uses content-based modeling with TF-IDF. It explains how KijiShopping collects user and product data, computes TF-IDF to find useful features, associates words with products using batch MapReduce jobs, determines a user's preferred words, and generates recommendations by combining user ratings and models using producers that access models via key-value stores. The goal is to provide real-time recommendations by leveraging the Apache Kiji framework for real-time analytics.