top of page
Keynote Speaker.
Razvan Dinu
Principal Applied Scientist NVIDIA
Razvan is a Principal Applied Scientist at NVIDIA focused on creating next-generation tools for building complex large-scale conversational systems. He is currently leading the engineering for NeMo Guardrails, an open-source toolkit for building safe LLM-powered conversational systems. Holding a Ph.D. in Artificial Intelligence, Razvan is also a seasoned full-stack software engineer with over 15 years of experience building technology-centric products. Over the past decade, he has founded multiple AI-focused startups where he acted as both CTO and CEO. During his studies Razvan has also won two silver medals at international Math Olympiads and 23 first prizes in national Math and Informatics contests in Romania.
Challenges in Building Safe
LLM-Powered Conversational Systems
Developing safe conversational systems powered by Large Language Models (LLMs) presents distinct challenges. This presentation focuses on the need to ensure dialogues stay on-topic, guarantee user safety, and maintain system security. We will touch on reducing model hallucinations, ensuring rigorous fact-checking, input/output moderation techniques, and strategies to deal with sensitive data. Additionally, we will briefly explore the intricacies of safety in multi-turn conversations. Finally, we will introduce the NeMo Guardrails toolkit by NVIDIA for building safer conversational systems.
Invited
bottom of page