AI Compiler and Agentic Optimization Engineer

About the role

We are looking for an AI Compiler and Agentic Optimization Engineer to build compiler technologies spanning model capture and graph import, intermediate representations, lowering and optimization passes, operator and kernel generation, and compiler-runtime integration.

This role is broader than kernel implementation alone. It also explores Agent/LLM-assisted optimization workflows for pass selection, kernel generation, search, performance diagnosis, and optimization planning, with strong correctness and reproducibility requirements.

What you will do

• Develop compiler front-end capabilities for model capture, graph import, operator decomposition, shape/type inference, and dynamic-shape handling.
• Design or extend IRs and dialects; implement lowering, legalization, canonicalization, graph transformation, and hardware-aware optimization passes.
• Generate and optimize high-performance operators and kernels using Triton, CUDA, or equivalent accelerator programming models, including fusion, tiling, layout, vectorization, and software pipelining.
• Develop scheduling, autotuning, and compiler-runtime coordination for asynchronous data movement and load-compute-store pipelines.
• Build Agent/LLM-assisted optimization workflows with automated correctness checks and performance validation.

What we are looking for

• Master’s degree or Ph.D. in Computer Science, Computer Engineering, AI, High-Performance Computing, or a related field.
• Representative work in at least one primary area: AI compiler front-end or IR, graph optimization and lowering, high-performance kernel generation, compiler scheduling/autotuning, or agentic compiler optimization.
• Hands-on experience with technologies such as MLIR, LLVM, TVM, XLA/StableHLO, PyTorch FX,
• TorchDynamo, Inductor, ONNX, or OpenAI Triton.
• Proficiency in C++ and Python, Linux, and CUDA/HIP/ROCm or equivalent NPU programming.
• Ability to demonstrate compiler correctness and measurable performance gains on realistic workloads.

Preferred qualifications

• Upstream contributions to compiler, kernel, or framework projects.
• Experience integrating compilers with production training or inference runtimes.
• Research or production experience in Agent/LLM-assisted code generation or compiler optimization.