Building the software backbone for AI processors that run on milliwatts, not megawatts.
Accelerator Compiler and Tool Chain Lead
We're rethinking how neural networks meet silicon—designing chips so efficient they can run sophisticated perception models for years on a coin-cell battery. This role owns the compiler and tool chain that transform high-level model descriptions into blazingly fast, ultra-low-power compute graphs on custom accelerator hardware. You'll define intermediate representations, build optimization passes, and work directly with silicon architects to expose hardware capabilities that most compilers never get to touch. If you've ever been frustrated by what existing ML frameworks leave on the table and want to close the gap between algorithm and transistor, this is the rare opening where you set the direction from scratch.
What they're looking for
- 10+ years writing production compilers or code generators, with deep experience in C/C++ and Python
- Fluency in at least one major ML framework (PyTorch, TensorFlow) and a strong grasp of how neural network operators map to hardware
- Track record of shipping tool chains for custom silicon—DSPs, NPUs, or domain-specific accelerators—not just off-the-shelf GPUs
- Comfortable reading RTL or microarchitecture specs and translating hardware constraints into compiler optimizations
- Proven ability to lead a small, senior team through ambiguous technical terrain without a playbook