Member of Technical Staff, GPU Optimization
MirageAbout the role
Mirage is the leading AI short-form video company. We’re building full-stack foundation models and products that redefine video creation, production and editing. Over 20 million creators and businesses use Mirage’s products to reach their full creative and commercial potential.
We are a rapidly growing team of ambitious, experienced, and devoted engineers, researchers, designers, marketers, and operators based in NYC. As an early member of our team, you’ll have an opportunity to have an outsized impact on our products and our company's culture.
Our Products
Our Technology
Press Coverage
Our Investors
We’re very fortunate to have some the best investors and entrepreneurs backing us, including Index Ventures, Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Uncommon Projects, Kevin Systrom, Mike Krieger, Lenny Rachitsky, Antoine Martin, Julie Zhuo, Ben Rubin, Jaren Glover, SVAngel, 20VC, Ludlow Ventures, Chapter One, and more.
** Please note that all of our roles will require you to be in-person at our NYC HQ (located in Union Square)
We do not work with third-party recruiting agencies, please do not contact us**
About the Role
As an expert in making AI models run fast—really fast—you live at the intersection of CUDA, PyTorch, and generative models, and get excited by the idea of squeezing every last bit of performance out of modern GPUs. You will have the opportunity to turn our cutting-edge video generation research into scalable, production-grade systems. From designing custom CUDA or Triton kernels to profiling distributed inference pipelines, you'll work across the full stack to make sure our models train and serve at peak performance.
Key Responsibilities
Optimize model training and inference pipelines, including data loading, preprocessing, checkpointing, and deployment, for throughput, latency, and memory efficiency on NVIDIA GPUs
Design, implement, and benchmark custom CUDA and Triton kernels for performance-critical operations
Integrate low-level optimizations into PyTorch-based codebases, including custom ops, low-precision formats, and TorchInductor passes
Profile and debug the entire stack—from kernel launches to multi-GPU I/O paths—using Nsight, nvprof, PyTorch Profiler, and custom tools
Work closely with colleagues to co-design model architectures and data pipelines that are hardware-friendly and maintain state-of-the-art quality
Stay on the cutting edge of GPU and compiler tech (e.g., Hopper features, CUDA Graphs, Triton, FlashAttention, and more) and evaluate their impact
Collaborate with infrastructure and backend experts to improve cluster orchestration, scaling strategies, and observability for large experiments
Provide clear, data-driven insights and trade-offs between performance, quality, and cost
Contribute to a culture of fast iteration, thoughtful profiling, and performance-centric design
Required Qualifications
Bachelor's degree in Computer Science, Electrical/Computer Engineering, or equivalent practical experience
3+ years of hands-on experience writing and optimizing CUDA kernels for production ML workloads
Deep understanding of GPU architecture: memory hierarchies, warp scheduling, tensor cores, register pressure,
Apply for this role
Generate a tailored application kit with a matched cover letter, interview prep, and CV highlights — in under 60 seconds.
Apply Now →Generate Application KitFree account required — sign up in 30s