The document discusses a learning and inference engine applied to a ubiquitous recommender system. It aims to help users access information by guiding them based on their context and situation. The system should be able to recommend information to help users achieve their goals. Major challenges include avoiding expert intervention, starting with no prior knowledge, quick learning, and adapting to changing user interests. The document presents a scenario where the recommender system infers relevant information to recommend to new employees without expert input, based on analyzing the actions of other employees in their teams. It aims to start with a predefined set of actions from social groups and progressively adapt recommendations to individual users.