Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn how Character.AI scales its massive AI entertainment platform serving tens of millions of daily users through this 32-minute conference talk from Ray Summit 2025. Discover how Haoran Li from Character AI built a fully open-source post-training stack using the Ray ecosystem to power their large language model infrastructure. Explore Rayman, Character AI's internal fine-tuning platform that accelerates model development velocity and significantly improves training efficiency for large Mixture-of-Experts (MoE) models by combining scalable distributed compute, efficient orchestration, and flexible experimentation using open-source components. Understand how Character AI leverages and adapts open-source reinforcement learning libraries like Verl to address unique RL challenges, enabling rapid iteration and high-quality post-training for LLM-based interactive characters. Examine the architecture of Rayman, the open-source projects forming the system's backbone, the RL framework driving model refinement, and key machine learning challenges overcome to deliver state-of-the-art AI entertainment at global scale. Gain practical insights into building scalable post-training pipelines, adapting reinforcement learning for large consumer applications, and accelerating model improvement using Ray-based infrastructure.
