We present the AutoConcierge system that can “understand” human dialogs in a specific domain, namely, restaurant recommendation. AutoConcierge will interactively understand a user’s utterances, and request the user to provide required information via a natural language reply. AutoConcierge uses GPT-3 to convert human dialogs into predicates that represent knowledge implicit in the dialogs. These predicates are then input into the goal-directed s(CASP) answer set programming (ASP) system for performing commonsense reasoning to compute responses in the form of predicates. GPT-3 is used again to convert these computed predicates into natural language sentences that are communicated to the user. To the best of our knowledge, AutoConcierge is the first automated conversational agent that can realistically converse like a human based on truly understanding user utterances. The framework used for AutoConcierge provides a recipe for developing other task-specific chatbots leveraging large language models and answer set programming.