Well-funded San Francisco AI startup making generative models that respond as fast as you can interact.
Kernel Engineer (ML Accelerators)
Referral bonus eligible. Know someone who would be great for this role? Refer them to a Fluency Digital recruiter. If they're placed, you may be eligible for a referral bonus of up to $5,000 (terms apply). Refer someone for this role
Every interactive AI product hits the same wall: the model is brilliant in a demo and unusable in the moment, because inference burns more time, memory, and power than the experience can afford. This company is betting that the fix lives below the framework, in hand-tuned kernels that squeeze every last cycle out of modern accelerators. As their kernel engineer, you'd own the performance floor for models that must generate output while the user is still watching, working in C/C++ and Python against PyTorch at the boundary where software meets silicon. It's a small team with serious backing, so the kernels you write this quarter ship to users the next.
What they're looking for
- 5+ years writing high-performance C/C++, with recent work on GPU or other accelerator kernels
- Deep grasp of memory hierarchies, throughput, and latency tradeoffs in ML inference workloads
- Fluent in Python and PyTorch internals, able to trace a slow forward pass down to the offending instruction
- Track record of measurable wins: faster kernels, lower memory footprints, or better hardware utilization you can quantify
- Comfortable on-site in San Francisco, collaborating closely with researchers and infrastructure engineers