AI Software Engineer, Lightspeed Studios

NVIDIA•Us, CA

12h

About The Position

Lightspeed Studios is seeking an AI Software Engineer to build and ship software for upcoming, not-yet-announced projects that bring world models and video diffusion models to real-time gaming. This role involves working from prototype to release, training, fine-tuning, and evaluating models, including data curation and adapting models to game-specific content and controls. The engineer will optimize models and inference for latency, token throughput, and quality using techniques such as distillation, quantization, and reduced-step sampling on RTX devices and in the cloud. Responsibilities also include profiling and removing bottlenecks across the stack, from model architecture to GPU kernels, and integrating models into Unreal Engine, Unity, and custom engines for efficient operation alongside rendering. This position requires leading technical decisions, mentoring other engineers, and collaborating with NVIDIA Research, rendering, and art teams.

Requirements

BS, MS, or PhD in Computer Science, Electrical Engineering, or a related field, or equivalent experience.
12+ years of software engineering experience, including substantial hands-on work shipping AI, deep learning, or machine learning systems.
Expert-level C++ and Python.
Hands-on experience training and fine-tuning deep learning models with PyTorch or a similar framework, including distributed training.
Experience in one or more of these areas: video generation, world models, diffusion or flow-matching models, transformers, or neural rendering.
Experience optimizing GPU inference with CUDA, TensorRT, TensorRT-LLM, or Triton.
A track record of turning complex prototypes into shipped products, and clear communication across research, engineering, and art.

Nice To Haves

Hands-on experience with interactive, action-conditioned, or real-time world models or video generation.
Experience developing custom CUDA or Triton kernels.
Experience with game engine development or real-time rendering pipelines.
Open-source contributions (such as Diffusers, FastVideo, vLLM, SGLang, or TensorRT-LLM), publications, or patents.

Responsibilities

Build and ship software for upcoming, not-yet-announced projects that bring world models and video diffusion models to real-time gaming, from prototype to release.
Train, fine-tune, and evaluate models, including data curation and adapting models to game-specific content and controls.
Optimize models and inference for latency, token throughput, and quality, using techniques such as distillation, quantization, and reduced-step sampling on RTX devices and in the cloud.
Profile and remove bottlenecks across the stack, from model architecture to GPU kernels.
Integrate models into Unreal Engine, Unity, and custom engines, so they run efficiently alongside rendering.
Lead technical decisions, mentor other engineers, and collaborate with NVIDIA Research, rendering, and art teams.