UST is hiring a Machine Learning Engineer (ML Engineer II) in Oregon (onsite). In this role, you will help design and optimize multi-agent workloads from the application layer down to inference runtime and accelerator hardware, with a focus on measurable performance improvements.
You’ll work on agent-centric capabilities such as agent orchestration, tool and function calling, memory planning, and agent-to-agent communication. If you enjoy profiling, benchmarking, and turning bottlenecks into faster, more efficient executions, this position offers a clear technical mission and the opportunity to influence both software and runtime behavior.
What you’ll do
- Develop a multi-agent workload at the application level, then understand, profile, and optimize execution through the full stack: inference runtime, OMIX/middleware, compute runtime, Linux GPU driver, and accelerator hardware.
- Optimize for latency, throughput, tokens/sec, memory footprint, and accelerator utilization.
- Select and tune models and inference engines based on agent workload characteristics, hardware capabilities, and performance requirements.
What you bring
- Hands-on development of multi-agent workloads.
- Experience with agent frameworks such as LangGraph/LangChain, AutoGen, and CrewAI.
- Strong understanding of LLMs, SLMs, and multimodal models, including Transformer architecture, attention, tokenization, context windows, and KV cache.
- Hands-on experience with model selection, evaluation, and deployment for different agentic workload requirements.
- Understanding of model formats and optimization, including ONNX, Open VINO IR, safe tensors, and quantization such as FP16/BF16/INT8/INT4.
- Experience with inference engines/runtime frameworks such as Open VINO, ONNX Runtime, vLLM, llama.cpp, or TGI.
- Knowledge of prefill vs. decode, batching, continuous batching, speculative decoding, KV-cache management, and memory optimization.
- Understanding of CPU/GPU model execution, device placement, and heterogeneous inference.
- Ability to benchmark and compare models, inference engines, and runtime configurations to identify performance bottlenecks.
- Understanding of GPU memory, kernel execution, synchronization, device selection, and host/device data movement.
- Familiarity with middleware/accelerator compute runtimes such as OMIX/OneAPI/SYCL, Level Zero, or OpenCL.
- Strong Python and working knowledge of C/C++.
- Experience with Git/GitHub, Linux shell, and debugging tools.
- Practical use of GitHub Copilot for development, debugging, and code generation.
- Docker/container fundamentals.
- Ability to benchmark and profile AI workloads, including understanding latency, throughput, tokens/sec, GPU utilization, memory bandwidth, and CPU/GPU bottlenecks.
- Ability to determine whether a performance issue originates in the agent, model, inference engine, runtime, or driver.
Compensation
$82,000 - $123,000 per year.
Benefits
- Full-time employees accrue a minimum of 10 days of paid vacation per year.
- 6 days of paid sick leave each year (pro-rated for new hires throughout the year).
- 10 paid holidays.
- Eligible for paid bereavement leave and jury duty.
- 401(k) Retirement Plan participation with employer matching.
- Medical, dental, and vision insurance for employees and their dependents residing in the US.
- Company-paid Employee Only benefits: basic life insurance, accidental death and disability insurance, and short- and long-term disability benefits.
- Regular employees may purchase additional voluntary short-term disability benefits.
- Eligible to participate in a Health Savings Account (HSA).
- Eligible to participate in a Flexible Spending Account (FSA) for healthcare, dependent child care, and/or commuting expenses.
Technologies: LangGraph, LangChain, AutoGen, CrewAI, Python, C/C++, Git, GitHub, GitHub Copilot, Docker, ONNX, Open VINO IR, safe tensors, FP16, BF16, INT8, INT4, Open VINO, ONNX Runtime, vLLM, llama.cpp, TGI, OMIX, OneAPI, SYCL, Level Zero, OpenCL.
Skills: Agentic AI, Python, Linux, MCP3, RAG, Kubernetes.
