Senior Research Scientist/Software Engineer, TikTok-Neural Graphics and World Models

About the Role We are looking for a Senior Research Scientist / Engineer to develop AI models, neural representations, data strategies, and evaluation methods for our neural graphics system. This is a hands-on research role for candidates who combine current ML expertise with depth in 3D or graphics domain. Depending on your primary focus track, you may define research directions and build AI models, data strategies, evaluation systems, or interactive prototypes across generative modeling, world models, 3D modeling and representations, animation, simulation, and rendering. Strong candidates combine research judgment with hands-on engineering ability: they can identify an important modeling problem, translate it into concrete research and engineering milestones, run rigorous experiments, and work with engineering teams to turn successful results into interactive, reliable engine capabilities. Job Responsibilities - Lead research directions in one or more areas of neural graphics, world models, 3D modeling, animation, simulation, or rendering. - Develop, train, adapt, and evaluate AI models and representations for interactive graphics and world-generation systems. - Define data and evaluation strategies that connect model behavior to quality, controllability, coherence, and interactive performance. - Build research prototypes and collaborate with engineering teams to bring successful models into real-time engine workflows. - Analyze results, identify technical risks, and translate research findings into clear next steps for model, data, and system development. - Produce reproducible research artifacts, communicate technical direction, and mentor team members. Minimum Qualifications - PhD/MS, or equivalent research experience in machine learning, computer graphics, robotics, applied mathematics, or a related technical field. - Relevant research or industry experience in neural graphics, generative AI, computer graphics, world models, or a closely related domain; doctoral research may count toward this experience. - Strong, up-to-date knowledge of modern machine learning, with deep expertise in at least one area of model architecture, training, data strategy, evaluation, or scaling and hands-on experience training or adapting generative models such as variational autoencoders, latent tokenizers, diffusion or flow-matching models, diffusion transformers, autoregressive models, or multimodal transformers. - Engineering fluency with ML frameworks such as PyTorch or JAX, with practical understanding of model development, representation learning, model scaling, distributed training, and pre-training and post-training tradeoffs. - Deep expertise in at least one of the following AI-for-graphics areas: 3D modeling, asset generation, or 3D representations; animation, motion generation, or character behavior; simulation, physical AI, or neural simulation; rendering or neural rendering; world models or video prediction; or graphics and game-engine workflows. - Demonstrated ability to formulate original hypotheses, design controlled experiments, and connect data composition, model behavior, and evaluation results to technical decisions. - Ability to collaborate with engineering teams to translate research prototypes into engine pipelines while preserving model quality, controllability, and interactive performance. Preferred Qualifications - Publications at leading research venues such as SIGGRAPH, CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, RSS, CoRL, or related top-tier conferences. - Experience creating novel model architectures, training recipes, data strategies, or evaluation methods for AI-for-graphics systems. - Experience with 3D or world generation conditioned on text, images, video, actions, or structured scene inputs, including controllability or multimodal alignment. - Experience with reinforcement learning, preference optimization, imitation learning, or model-based control for interactive world models, where relevant to the focus track. - Experience moving models into real-time graphics or engine workflows, including distributed training or inference optimization for interactive use.