Machine Learning Compiler Engineer

Jane Street•New York, NY

About The Position

Jane Street is looking for compiler engineers to join its new ML Compilers Team. Machine learning is a critical part of the business, and the firm's trading environment provides a rapid-feedback platform for ML experimentation. The ML team values software engineering craft and the design of useful, well-crafted APIs and systems. ML compilers help turn research ideas and novel architectures into high-throughput training jobs and low-latency inference. The team works to understand and improve existing ML compiler technologies and determine where to build its own solutions, with the goal of making effective use of hardware resources and enabling rapid model development and deployment. In this role, the engineer would shape the development of the compiler stack.

Requirements

Experience working on a machine learning compiler is ideal; non-ML compiler experience is also valuable.
Experience with any part of the compiler stack is relevant, including ML frameworks such as PyTorch, JAX, or TensorFlow; graph compilers such as Inductor, XLA, or TVM; or kernel DSLs such as Triton, Helion, or cuTile.
Understand how the different parts of the ML compiler stack fit together, even if unfamiliar with specific parts.

Nice To Haves

Experience with ML compiler optimizations such as algebraic rewriting, materialize-versus-recompute tradeoffs, fusion, equality saturation, scheduling languages, loop-level optimization, or autotuning.
Experience with ML compiler frontends, including graph capture, symbolic and dynamic shapes, control flow, or automatic differentiation.
GPU kernel programming and architecture knowledge, including Triton, cuTile, CuTeDSL, CUDA, compute and memory hierarchies, or programming with tensor cores.
Knowledge of compiler concepts such as static analysis, intermediate representations including SSA and CFGs, or instruction selection.
Knowledge of ML inference techniques such as CUDA Graphs, KV caching, or batching.
Knowledge of ML training techniques such as Tensor, Sequence, Expert, or Pipeline Parallelism; FSDP; NCCL; or activation checkpointing.
Experience with GPU debugging and optimization tools such as CUDA-GDB, Nsight Systems, or Nsight Compute.
Interest or experience in formal methods, including proof assistants, automated theorem provers, or agentic proof systems.
Knowledge of programming languages and DSL design, including embedded DSLs, staging and metaprogramming, type systems, or functional programming.