The role
- COMP
- $200K - $200K
- EQUITY
- 0.5 - 1%
- LOCATION
- San Francisco
- WORKPLACE
- On-site
- EXPERIENCE
- 1 - 5 years
- VISA
- None, Visa transfers
- STACK
- Python, AWS, C/C++, CUDA, HIP, Triton, vLLM-class serving, Kubernetes
- INDUSTRY
- AI, Hardware
/roles — ROLE_72
Member of Technical Staff
A company that builds AI agents that work as autonomous performance engineers, optimizing GPU kernels for AI inference
The company
AI-infrastructure company building the fastest inference for open models in the enterprise.
- STAGE
- seed-stage
- FUNDING
- seed funding
- TEAM
- founding team of <10
- FOUNDED
- 2025
JD — the work
About the role
Maximize intelligence per watt: serve open-source LLM inference at the best performance per dollar through autonomous optimization of heterogeneous hardware — NVIDIA, AMD, TPU, Trainium and beyond. Small SF team, complete autonomy, kernels to customers.
What you'll do
- Ship day-zero support for new open models, tuned for latency and throughput
- Optimize serving: batching, KV cache, speculative decoding, quantization
- Write and tune kernels in CUDA, HIP, and Triton across vendors
- Design and operate heterogeneous clusters
- Run production inference at scale: reliability, observability, cost per token
What they're looking for
- Deep systems/GPU engineering ability
- First-principles problem solving with unreasonable standards
- San Francisco, on-site 5 days/week