Senior/Staff Software Engineer - LLM Inference Systems & Open-Source

Qualcomm•San Diego, CA

2d•$162,000 - $243,000

About The Position

We are looking for experienced software engineers to help build and advance the software ecosystem for large-scale LLM inference on Qualcomm platforms. In this role, you will work at the intersection of open-source LLM serving systems, distributed systems, cloud-native infrastructure, and AI accelerator software. You will contribute to open-source projects such as llm-d and related LLM inference engines, while developing the integrations and capabilities required to run these systems efficiently and reliably on Qualcomm AI accelerators. Depending on your background and interests, you may focus more heavily on either distributed LLM serving and orchestration or inference runtime and accelerator integration, while collaborating across both areas. This is an opportunity to work directly with leading open-source communities and help shape the next generation of scalable, production-grade LLM inference infrastructure.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, Electrical Engineering, or a related technical field, or equivalent practical experience.
Strong software engineering and debugging skills.
Experience developing production-quality software in one or more of Python, C++, Go, or Rust.
Strong understanding of at least one of the following areas: Machine learning or LLM inference systems, Distributed systems, AI accelerator or GPU software, Cloud-native/Kubernetes infrastructure, ML compiler or runtime systems.
Experience working with complex software systems and debugging problems across multiple layers of the stack.
Ability to collaborate effectively across teams and communicate technical concepts clearly.

Nice To Haves

Experience in one or more of the following areas is highly desirable: LLM inference engines, model-serving frameworks, or distributed inference systems.
Architecture and implementation of high-performance LLM inference runtimes, including model execution, scheduling, memory management, attention, and KV-cache management.
Distributed LLM inference architectures, including prefill/decode disaggregation and KV-cache transfer.
llm-d, KServe, Kubernetes Gateway API, Kubernetes operators, or cloud-native AI infrastructure.
PyTorch internals, torch.compile, Dynamo, Inductor, Triton, MLIR, or other compiler/runtime technologies.
Hardware accelerator software stacks, GPU programming, or custom AI accelerator integration.
RDMA, RoCE, GPU Direct-style communication, PCIe, high-speed networking, or distributed communication libraries.
Model parallelism, including tensor parallelism, pipeline parallelism, expert parallelism, and data parallelism.
Performance profiling and optimization of distributed AI workloads.
Kubernetes scheduling, autoscaling, networking, observability, and production operations.
Open-source development, including contributing patches, participating in design discussions, reviewing code, and collaborating with upstream maintainers.
Experience working across hardware/software boundaries with hardware, firmware, compiler, runtime, and systems teams.

Responsibilities

Contribute features, improvements, bug fixes, and architectural enhancements to open-source LLM inference projects, including llm-d and other open-source LLM inference and serving projects.
Develop and improve support for Qualcomm AI accelerators in open-source LLM serving frameworks.
Work with upstream maintainers and open-source communities to design, review, and upstream Qualcomm-related functionality.
Develop and enhance inference engine integrations for Qualcomm platforms, including device abstraction, accelerator runtime integration, memory management, distributed execution, compilation, and performance optimization.
Develop and improve llm-d integration and deployment capabilities for Qualcomm platforms, including reference architectures, deployment blueprints, configuration, and operational tooling.
Design and implement distributed LLM inference capabilities such as: Prefill/decode disaggregation, KV-cache management and transfer, Distributed communication and high-performance data movement, Request routing and load balancing, Tensor, pipeline, expert, and other forms of model parallelism, Multi-node and scale-out inference, Autoscaling and resource management, Failure detection, recovery, and service resiliency.
Integrate LLM inference software with Kubernetes and cloud-native infrastructure, including scheduling, autoscaling, networking, observability, and lifecycle management.
Profile, benchmark, and optimize end-to-end inference performance, including time to first token (TTFT), inter-token latency (ITL), throughput, memory utilization, and scalability.
Diagnose complex issues spanning model-serving frameworks, distributed runtimes, operating systems, networking, accelerator runtimes, and hardware.
Collaborate closely with hardware, firmware, compiler, runtime, Kubernetes, and product teams to define requirements and deliver end-to-end solutions.
Develop technical documentation, examples, reference deployments, and best practices to help customers and developers deploy LLM workloads on Qualcomm platforms.
Participate in architecture discussions and help influence the technical direction of Qualcomm's open-source LLM inference software strategy.