AI Journal Club ft. Google + NVIDIA

Join Workato's AI Journal Club series—we're bringing together the best AI researchers to share papers and exchange perspectives on how AI research is shaping real-world systems. Who Should Attend AI Researchers and practitioners working at the intersection of AI research and real-world systems. Schedule 5:30–6:00 PM: Check-in and registration6:00–6:15 PM: Welcome to Workato  6:15–6:45 PM: Talk by Allen Chuang (Google)6:45–7:00 PM: Q&A7:00–7:30 PM: Talk by Ray Liu (NVIDIA)7:30–7:45 PM: Q&A7:45-8:30 PM: Networking Please arrive by 6:00 PM. We politely ask that attendees arrive by this time out of respect for our speaker. Sessions Allen Chuang — Taming the Dynamics of LLM Reasoning: Test-Time Exploration, Data Scheduling, and On-Policy Distillation Scaling reasoning capabilities in Large Language Models (LLMs) requires navigating complex optimization and search dynamics across both inference and post-training. However, modern reasoning pipelines suffer from critical inefficiencies throughout the model lifecycle: test-time decoding wastes compute on redundant search trajectories, reinforcement learning (RL) struggles with uniform data pacing, and on-policy distillation frequently undergoes catastrophic truncation collapse driven by length inflation. In this talk, I will present a unified perspective on understanding and taming reasoning dynamics. First, we examine inference-time search, demonstrating how Decoding Tree Sketching (DTS) enables structured exploration and early termination at decision tokens to eliminate redundant rollouts without retraining. Next, we turn to post-training optimization, showing how Adaptive Data Scheduling (ADS) leverages semantic clustering and policy-boundary selection to dynamically pace LLM RL curricula. Finally, we analyze the failure modes of On-Policy Distillation and introduce StableOPD to suppress trajectory explosion and restore distillation stability. Together, these methods provide practical mechanisms for building efficient, robust, and scalable LLM reasoning pipelines. Ray Liu — Nondeterminism in LLM Inference & Training–Rollout Mismatch LLM generation is not deterministic even when the temperature is set to zero. System-level configuration changes, such as variations in batch size and parallel strategy, which commonly occur in real-world serving due to continuous batching. This issue is more pronounced in RL, where the training and rollout engines naturally operate with different batch sizes, kernel selections, and parallelization strategies. This training-rollout mismatch problem leads to suboptimal performance and training collapse, specifically for the MoE model. In this talk, he will analyze why this happens and how to solve the problem at the system level by building deterministic GPU kernels. Featured Speakers Allen Chuang — Yu-Neng (Allen) Chuang is a Research Scientist at Google DeepMind. He works on building reliable and efficient LLM agentic systems through continued pre-training and post-training. He received his Ph.D. in Computer Science from Rice University. His research focuses on LLM reasoning, post-training, and agentic systems, with the goal of developing scalable and reliable AI systems for real-world applications. His work has been published at leading AI and machine learning venues, including ICML, ICLR, and NeurIPS, and has received several recognitions, including an ICML Spotlight, a CIKM Best Demo Paper Honorable Mention, and a NAACL Best Paper Award nomination. Ray Liu — Zirui (Ray) Liu is an Assistant Professor of Computer Science at the University of Minnesota. His interests lie in the broad area of LLM and MLSys. He regularly published papers in top venues such as NeurIPS, ICML, ICLR, and MLSys. His work has been integrated into widely used NLP tools like Llama.cpp, SGLang [in progress], and Huggingface Transformers, and was highlighted at Google I/O sessions. Host​About Workato Workato is the Enterprise MCP company, providing the connective lay…