Principal Software Architect

I’m working with a well-funded AI compute company building a programmable accelerator platform and the complete software stack required to run modern AI workloads efficiently.

They are hiring a senior architect to shape the full path from models and frameworks through graph capture, compiler, runtime and driver layers, down to kernels and hardware execution. This is a rare role for someone who can connect high-level workload behaviour with low-level GPU architecture and code generation.

You will work across:

• PyTorch, JAX and ONNX integration
• Graph capture, partitioning and optimization
• MLIR/LLVM lowering and backend code generation
• Runtime, scheduling and driver architecture
• SIMT execution, occupancy and memory hierarchy
• CUDA and Triton programming models
• Kernel fusion, tiling and custom operator lowering
• Hardware/software co-design and platform validation
• Performance profiling from models down to instructions

They are looking for:

• 12+ years across AI, compiler, runtime or systems software
• Strong C++ with deep LLVM and MLIR experience
• Expertise in GPU/SIMT execution and CUDA programming
• Understanding of modern AI models at graph and operator level
• Experience connecting frameworks, compilers, runtimes and hardware
• A hands-on, measurement-led approach to architecture
• Experience guiding senior engineers through design and code reviews

Experience with custom accelerators, PyTorch compilation, XLA, Triton, quantization, performance modelling or pre-silicon software development would be particularly valuable.

This is an opportunity to define the software architecture for a new compute platform as it moves from a working stack into large-scale production.