ML Compiler & System
Job description
Owns the Quadrants open-source Python-to-machine-code translation layer end to end, covering the API, IR and lowering, optimization passes, numerical primitives, and roadmap. The role is part research and part engineering, focused on making performance-critical numerical Python fast across CPU and GPU backends.
At a glance
- Work mode
- Remote
- Languages
- English
What you'll do
- Own the API and authoring experience: the surface physics and numerical-computing engineers write against, idiomatic and natural to Python users, with clean decorators, legible constraints, sharp errors, and real escape hatches
- Own the IR, lowering and backends: a stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation
- Own optimization: turn naive code into strong execution plans through vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, and dropping needless synchronization
- Own primitives such as Cholesky, eigendecomposition, matmul, reductions, and solver kernels, including correctness, benchmarks, and regression tracking
- Watch Genesis and the compiler together, hunting for performance wins on both sides and proposing API changes that cut duplication, improve readability, and reduce platform-specific special-casing
- Treat the project as a product: benchmark real workloads, track performance over time, catch regressions early, stay close to users (Genesis first), find friction and missing abstractions, drive adoption, and set direction
What we're looking for
- Reason about whole systems and have taste for where complexity should live
- Measure before optimizing, reaching for memory traffic and launch overhead before FLOPs
- Have opinions about API design
- Write tests and benchmarks as you go, not afterward
- GPU knowledge: what optimal looks like and why, including access patterns, registers, shared memory, barriers, atomics, and threading
- Idiomatic Python and API taste: a feel for what's natural to Python users and APIs that free users rather than constrain them
- Compiler knowledge: the path from authoring to kernels (IR, lowering, optimization), weighed against maintainability
- Proactive and largely self-directed; fetch context across the company instead of waiting for it
- Comfortable talking to the engineers who use what you build and at ease designing in the open for users you'll never meet
- Ready to grow into leading a small team (a couple of people, near term)
Nice to have
- CPU intuition
Your team
Quadrants
An open-source translation layer that turns ordinary Python into optimized machine code at runtime across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends; a standalone project with its own roadmap, built in the open, with Genesis as its first and top-priority user.
Apply
Direct apply at Genesis
Opens the job posting at Genesis
About Genesis
Genesis builds a physics simulator; Quadrants is its first and top-priority user, the source of its hardest workloads and benchmarks.
Key facts
- Quadrants is a standalone open-source project with its own roadmap, built in the open
- Genesis is the first and top-priority user of Quadrants
- The ambition is to build the best Python compiler in the world for physics and performance-critical numerical computing
- They intend to beat leading alternatives like Nvidia Warp, Numba, or Triton
What they value
- Usability comes first; when usability and performance collide, usability wins unless the cost is severe
- Performance is the compiler's job, not the user's
- High-performance code should feel natural to write
- Understanding is real work, not overhead
- Treat nothing you did not write for granted