The role
- COMP
- $180K - $230K
- EQUITY
- Competitive
- LOCATION
- South Bay Area
- WORKPLACE
- On-site
- EXPERIENCE
- 1+ years
- VISA
- None, Visa transfers, New visa sponsorships
- STACK
- PyTorch
- INDUSTRY
- AI, Devtools, Robotics, API SDK
/roles — ROLE_317
Member of Technical Staff, ML Systems
Seed-stage team making world-model and video generation run far faster and cheaper on GPUs
The company
Seed-stage startup rebuilding AI compute software for world-model, video, and image workloads, spanning GPU kernels to distributed serving, with customers in generative media and robotics.
- STAGE
- Series A-stage
- FUNDING
- $10M+ raised
JD — the work
About the role
You would be responsible for making every layer of a GPU compute stack, built for world-model and video and image generation, fast and cheap to run, from hand-tuned kernels up to multi-node inference and training systems. The company is a small, in-person founding team that recently emerged from stealth with public benchmarks and a production API, and you would work directly with founders whose experience covers research, cloud infrastructure, kernel work, and distributed systems. It suits someone more excited by making a video model run an order of magnitude faster than by training it.
What you'll do
- Push GPU efficiency for training and serving world-model, video, and image workloads
- Profile kernels, memory, systems, and whole clusters with Nsight and similar tools to remove bottlenecks
- Write CUDA and Triton optimizations that ship on production code paths
- Build distributed inference and training engines for diffusion models spanning many GPUs and nodes
- Own communication speed across NCCL, RDMA on InfiniBand or RoCE fabrics, and disaggregated serving setups
- Set up benchmarks and regression tests that keep performance gains from eroding in production
What they're looking for
- 1+ years of hands-on GPU performance or ML systems engineering
- Working knowledge of CUDA or Triton and GPU profiling tools
- Experience with PyTorch and distributed training or inference across multiple nodes
- Familiarity with GPU communication such as NCCL and RDMA networking
- Willingness to work in person with a small founding team in the Bay Area