Agentic AI / Multi-Agent Systems Engineer
$72,000 - $108,000 per year
Apply (opens in a new tab)Agentic AI / Multi-Agent Systems Engineer
Posted yesterday
SENIORITY
Senior
SALARY
$72,000 - $108,000 per year
About the role
- Born digital, UST transforms lives through the power of technology. We walk alongside our clients and partners, embedding innovation and agility into everything they do. We help them create transformative experiences and human-centered solutions for a better world.
- UST is a mission-driven group of 29,000+ practical problem solvers and creative thinkers in more than 30 countries. Our entrepreneurial teams are empowered to innovate, act nimbly, and create a lasting and sustainable impact for our clients, their customers, and the communities in which we live.
- With us, you'll create a boundless impact that transforms your career-and the lives of people across the world.
- Visit us at UST.com.
- You Are:
- UST is searching for an Agentic AI / Multi-Agent Systems Engineer who will develop and optimize multi-agent AI systems and agentic workloads.
- Work with agent frameworks and orchestration technologies to design scalable solutions
- Select, evaluate, and deploy appropriate models and inference engines based on workload requirements
- Debug and optimize performance across the full stack: agent model inference engine hardware
- Benchmark and profile AI workloads to identify and resolve performance bottlenecks
- Optimize model execution for latency, throughput, memory efficiency and accelerator utilization
- Collaborate on the complete execution path from agent logic through GPU hardware
- Bachelor's degree in Electrical Engineering or Computer Engineering 2-4 years of platform debug experience with BIOS/FW/platform ingredients
- Required Experience & Knowledge
- Linux & Systems
- Strong Linux development and debugging skills
- Understanding of processes, threads, memory, PCIe, DMA, kernel modules, and device drivers
- Ability to debug workloads across the application runtime driver stack
- Agentic AI & Multi-Agent Systems
- Hands-on development of multi-agent workloads
- Experience with agent orchestration, tool/function calling, memory planning and agent-to-agent communication
- Experience with agent frameworks such as LangGraph/LangChain, AutoGen, or CrewAIModels & Inference
- Strong understanding of LLMs, SLMs and multimodal models
- Knowledge of Transformer architecture, attention, tokenization, context windows, and KV cache
- Hands-on experience with model selection, evaluation, and deployment
- Understanding of model formats and optimization: ONNX, Open VINO IR, safe tensors, quantization (FP16/BF16/INT8/INT4)
- Experience with inference engines: Open VINO, ONNX Runtime, vLLM, llama.cpp, or TGIUnderstanding of prefill vs. decode, batching, continuous batching, speculative decoding, and KV-cache management
- Ability to optimize for latency, throughput, tokens/sec, memory footprint, and accelerator utilization
- Accelerator & Compute Stack
- Understanding of the complete execution path: Agent Model Inference Engine Middleware Compute Runtime GPU Driver Hardware
- Understanding of GPU memory, kernel execution, synchronization, device selection and host/device data movement
- Familiarity with middleware/accelerator compute runtimes: OMZ/OneAPI/SYCL, Level Zero, or OpenCLProgramming & Development
- Strong Python skills
- Working knowledge of C/C++Git/GitHub and Linux shell proficiency
- Debugging tools expertise
- Practical use of GitHub Copilot for development, debugging and code generation
- Docker/container fundamentals
- Performance Engineering
- Ability to benchmark and profile AI workloads
- Understanding of latency, throughput, tokens/sec, GPU utilization, memory bandwidth and CPU/GPU bottlenecks
- Ability to identify performance issues at different stack levels
- Intel GPU architecture and Linux GPU driver stackDRM/i915 and newer Intel GPU driver architecture
- Intel GPU profiling and telemetry
- Speculative decoding and continuous batching experience
- Tensor/pipeline parallelism knowledge
- Model conversion, graph optimization and operator/kernel fusionMCP (Model Context Protocol)
- Advanced RAG architectures
- Agent observability/tracing
- Model routing and dynamic model selection
- Distributed/multi-agent executionCUDA/HIP/ROCm familiarity
- Kubernetes, Ray or Slurm experience
- Distributed inference knowledge
- Linux kernel/GPU debuggingperf, gdb, tracing and GPU profiling tools
Before you apply
Applying takes about a minute. These four things decide how fast it moves after that.
Your profile is current
It's what we read first. Occupations, seniority and locations matter more than a long history.
Two examples you can talk through
Not a portfolio — just two pieces of work where you can explain the decisions and what you'd change.
A number in mind
What you're on now and what would make you move. We negotiate better when we know both.
Your notice period
Employers plan around it, and it's the question that stalls offers most often.
Once you apply, someone reads it and calls you before anything reaches the employer — usually within two working days.
