Diffusion-based LLM startup making AI radically faster and cheaper.
Machine Learning Systems Engineer
You'll spend your days squeezing inference latency out of diffusion models—rewriting kernels, profiling memory bandwidth, and deciding when a custom CUDA op beats PyTorch's defaults. One morning might mean benchmarking a new attention variant on H100s; the afternoon, debugging why a quantized checkpoint OOMs at batch size eight. This is work where shaving milliseconds translates directly to dollars saved for customers running million-token generations.
What they're looking for
- 2–5 years shipping ML infrastructure in production, with solid footing in Python and at least one systems language (C/C++, Rust, or Go)
- Hands-on experience optimizing deep learning training or inference—kernel fusion, custom operators, memory planning, or distributed strategies you've actually debugged
- Comfortable navigating cloud GPU stacks: you've set up clusters on AWS, GCP, or Azure, and have opinions on Kubernetes schedulers for ML workloads
- Track record with at least one major framework (PyTorch or TensorFlow) including lower-level APIs like torch.fx, torch.compile, or XLA
- Genuine curiosity about diffusion architectures—you've read the papers, maybe reimplemented a sampler, and get why consistency models matter for throughput