Chatbots are hard. They’re either safe, scripted, and boring, or unscripted and dangerously open. LLMs open up new opportunities for the industry to finally design rich chatbot experiences, but they lack true world models and robust guardrails. For the past five months, the Entertainment Technology Center at USC has been working on an open-source solution to this problem. With financial support from USC’s research fund, as well as Warner Bros and Microsoft, Project Ori is an attempt to build a chatbot architecture capable of explicitly referencing (or staying away from) certain topics, emotions, and moral domains. This presentation will explore how Project Ori leverages knowledge graphs, embeddings, and retrieval-assisted generation to empower chatbot designers. By integrating topic modeling and emotional tonality detection, Project Ori enhances the capability of chatbots to understand and appropriately respond to a wide range of topics and sentiments. Attendees will learn about the innovative methodologies behind this project, its practical applications, and see a live demonstration of its functionalities.