Senior Research Engineer, Interactive World Models

NVIDIA•Santa Clara, CA

1d•$184,000 - $356,500

About The Position

FlashDreams and FastGen are NVIDIA’s core technologies for turning video models into real-time world simulations. The stack spans model adaptation for faster generation and richer control, plus the execution layer that runs those models as responsive experiences. We develop and ship this technology in pursuit of generative worlds that people can explore and direct. The work can enable autonomous-driving simulation, robot policy development and testing, game worlds, medical simulation, and virtual training. As a Senior Research Engineer, you will lead engineering across model development and runtime systems, building capabilities and turning research into systems that work in real applications. We work across the world-model ecosystem, from emerging startups to established model labs. If you want to collaborate with leading researchers, shape a new computing platform, and ship AI capabilities with real-world impact, we would love to hear from you.

Requirements

Experience in one or more areas such as video or world models, diffusion and generative modeling, model distillation and adaptation, simulation, robotics, computer vision, or real-time, stateful ML systems.
MS or PhD in Computer Science, Electrical Engineering, or a related field (or equivalent experience).
5+ years of equivalent experience in applied ML or research engineering.
A record of advancing applied ML or ML systems through research, open-source software, patents, or deployed technology, including taking ambiguous ideas through thorough evaluation and release.
Hands-on experience with Python, PyTorch and GPU-accelerated training, inference, performance optimization, or serving.
Research and engineering judgment, with the ability to find practical solutions to open-ended problems and deliver reliable results within software and hardware constraints.

Nice To Haves

Experience with post-training generative video models, including distillation, self-forcing, action conditioning, or long-horizon memory.
Experience building and optimizing real-time, stateful generative inference systems, including history and KV-cache management, GPU kernels, quantization, parallel execution, streaming, scheduling, or multi-user serving.
Technical stewardship of an open-source ML project used by researchers or developers, with work spanning its architecture, APIs, model ecosystem, releases, maintainer practices, documentation, benchmarks, adoption, or community. Relevant ecosystems include vLLM, SGLang, FastVideo, LightX2V, Diffusers, FlashInfer, and comparable platforms.

Responsibilities

Build and optimize the continuous autoregressive serving loop, including per-step control inputs, model and KV-cache state management, GPU inference, frame streaming, and model integrations to speed-of-light.
Advance the production-ready world model frontier by working with researchers on few-step distillation, causal or autoregressive generation, reward fine-tuning, action conditioning, and long-horizon spatiotemporal memory and consistency.
Lead end-to-end delivery of capabilities such as multi-user experiences and simulation workflows, from prototype through evaluation, integration, and release. Partner with applied researchers and domain teams to meet quality, performance, and reliability goals.