Research Roundtable by Troveo : From Cognition to Embodiment

​Research Roundtable: From Cognition to Embodiment

​Tuesday, March 10, 2026 | 5:30 PM – 8:30 PM
EF's Office, San Francisco

​What does it take to build AI that thinks, moves, and sees in the real world? This roundtable brings together five researchers working at the sharpest edges of the field — from mathematical frameworks for AI safety, to real-time video generation, to training small models that beat frontier systems on embodied reasoning. Expect candid talks, honest tradeoffs, and the kind of depth you don't get from a product demo.

​Sponsored by Troveo This event is made possible with support from Troveo — the platform building premium, rights-cleared video datasets for the world's top AI labs. From embodied agent training data to avatar development and next-gen video models, Troveo delivers production-ready datasets at scale. Learn more at troveo.ai.

​

​The Lineup

​Karthik Ragunath Ananda Kumar — Sr. AI Researcher, Tavus Karthik is a Senior AI Researcher at Tavus, where he works on the Phoenix model family for real-time conversational avatars. He specializes in flow matching, diffusion models, and 3D Gaussian Splatting to achieve sub-100ms latency video generation. His talk will cover state-of-the-art Diffusion and Flow-Matching modeling architectures — a deep dive into the cutting edge of generative video AI.

​Sanjayan Sreekala — Applied AI Researcher, eBay | NeurIPS 2025 EAI Challenge Winner Sanjayan is an Applied AI Researcher at eBay and winner of the NeurIPS 2025 Embodied Agent Interface Challenge. His talk, "Winning the Embodied Agent Interface Challenge — What Worked and What Didn't", breaks down what it actually took to reach first place: fine-tuning small LLMs to outperform frontier models on embodied reasoning, the scrappy strategies that pushed him over the edge, and his perspective on where foundation models and robotics are headed next.

​Haoyi Niu — PhD Researcher, BAIR (UC Berkeley) Haoyi is a PhD student at BAIR, UC Berkeley, advised by Prof. Koushil Sreenath. He holds a B.S. and M.S. from the Department of Automation at Tsinghua University, where he also worked at the AIR-DREAM Lab under Prof. Xianyuan Zhan. His research spans reinforcement learning, long-horizon planning, and sim-to-real transfer for embodied agents, with publications at NeurIPS, ICML, ICRA, and ICLR.

​Bryce Allen Bagley, MD, MS — AI/ML Researcher, Stanford Bryce is a physician-scientist whose training spans physics, computer science, medicine, and applied mathematics. As a medical student at Stanford, he led an AI/ML research group within the Department of Neurosurgery and taught graduate coursework on mathematical and AI methods in medicine. His talk, "From Bits to Cogits: A Mathematical Bridge Between AI Safety and Cognitive Science", introduces research extending information theory to the quantification of human cognition — building a mathematically rigorous framework for AI alignment and the study of manipulation in human-AI interaction.

​Erwann Millon — Founding Engineer, Krea.ai Erwann is one of the earliest engineers at Krea.ai, where he built the infrastructure for serving hundreds of generative AI models at scale across multi-cloud GPU clusters. His expertise spans MLOps, deep learning inference optimization, and diffusion-based image and video generation. His talk will dive into real-time video generation — covering the engineering and modeling challenges behind making generative video fast enough to be truly interactive, drawing from his work pushing the boundaries of AI-powered creative tools at Krea.

​Agenda

​5:30 PM — Networking + light appetizers
6:00 PM — Talks + Q&A
7:30PM — Discussion + Networking

​Capacity strictly limited. Application required.

​Venue

​Entrepreneurs First (EF) turns exceptional individuals into funded founders. Some might have an early idea, others may need a cofounder. What matters is their potential. We back these founders first, then give them a bridge to Silicon Valley. To date, their companies are worth > $15B.

​​The AI Collective is a global non-profit building the human layer for the AI age. We unite 150,000+ leaders, builders, and stakeholders across 100+ forums worldwide to democratize the frontier, build trust, and coordinate how society navigates the rapid acceleration of technological progress.

​​All attendees and organizers at events affiliated with The AI Collective agree to our privacy policy and are subject to our code of conduct.