This document provides guidance on designing recommendation engine architectures. It discusses collecting user data through instrumentation, learning from user interactions via A/B testing, and determining the context and mechanism of recommendations. The document states that recommendation engines are essentially classification systems and outlines key components of recommendations like familiarity, trust, formality of suggestions, and efficacy that are analogues to social interactions. It recommends considering how to attract and engage users, present recommendations, collect requisite data, learn from users and performance to improve recommendations.