Agenda

Krafton Deep Learning engineers and NVIDIA experts deep dive into PUBG: Ally, a real-time AI companion that talks, jokes, reacts to player commands and brings extra fun and personality to every match. This session details the implementation of a complete conversational pipeline—spanning speech-to-text, small language model, and text-to-speech —powered by NVIDIA ACE. Attendees will learn how cutting-edge models were integrated and explore the latency reduction techniques required for seamless combat interaction.

Takeaway

Attendees will gain a comprehensive understanding of the PUBG: Ally architecture, examining how the full ASR, SLM, and TTS pipeline functions in a live environment. The session will provide technical details on optimizing inference performance using NVIDIA Nemotron, while demonstrating practical methodologies for calculating and reducing latency metrics to ensure seamless player interaction.

Intended Audience

AI and Gameplay Programmers interested in integrating Generative AI into real-time game loops.

Technical Directors evaluating architectures for on-device inference.

ML Engineers looking to optimize ASR, SLM, and TTS pipelines for low-latency interactive applications.