Job ID 2026-19339 Date posted Sep. 09, 2026 Location Seattle, Washington Category Software Engineering
As a Software Engineer on our AI Inference Runtime team, you will set technical direction for critical components of distributed Inference runtime for running SOTA AI Models.
You will lead hands-on work across scheduling, batching, KV-cache management, memory allocation, distributed workload execution, kernel development and optimization, and performance benchmarking and analysis. Your work will directly influence how efficiently new models use available compute. Partnering with our AI Infrastructure, compute, and product teams to enhance the performance and efficiency of Arm’s AI platform.
Responsibilities:
- Define the architecture, interfaces, and roadmap for AI inference runtime capabilities, including abstractions that support evolving models, workloads, and compute platforms.
- Enable new model architectures end to end through operator support, production validation, and optimization of scheduling, batching, model execution, memory management, and KV-cache efficiency.
- Profile system bottlenecks and develop optimized kernels and data-movement paths across compute, memory, networking, and framework integration.
- Evaluate new inference techniques and build benchmarking, regression, validation, and safe-rollout systems to improve latency, throughput, reliability, and resource efficiency.
- Partner with cloud, framework, compiler, hardware, and research teams; lead technical reviews, mentor engineers, and establish meticulous performance-engineering practices.
Required Skills and Experience :
- 5+ years of experience, or equivalent demonstrated impact, in ML systems, high-performance systems, compilers, kernel development, or production AI inference.
- Deep understanding of modern AI inference, including model execution, Attention, MoE, batching, prioritisation, and KV-cache behavior.
- Strong programming skills in C++, Rust, Python, or a comparable language, with knowledge of concurrency, parallel programming, handling of memory resources, and data movement.
- Proven ability to profile, debug, and optimize performance across kernels, runtimes, frameworks, operating systems, and hardware.
“Nice To Have” Skills and Experience :
- Experience developing or modifying inference schedulers, cache managers, batching systems, disaggregated or distributed execution paths.
- Experience optimizing kernels using accelerator programming tools, assembly, or intrinsics, including attention, matrix multiplication, operator fusion, and low-precision execution.
- Familiarity with model parallelism, collective communication, high-performance networking, compilers, or graph optimization.