The ideal candidate combines strong systems experience with hands-on experience in measurement, benchmarking, and performance analysis of modern ML systems.

What You’ll Do:

• Build simulation models for compute, memory, interconnect, and communication behavior in ML systems.

• Develop tools to simulate performance for training and inference workloads.

• Model distributed execution across accelerators, hosts, and network fabrics, including collectives, synchronization, and communication bottlenecks.

• Use simulation and analytical modelling to evaluate tradeoffs, identify bottlenecks, and guide system design.

• Run performance experiments and benchmarks on real ML systems to calibrate and validate simulation models.

• Analyze end-to-end performance, including throughput, latency, scaling efficiency, utilization, and cost/performance tradeoffs.

• Partner with hardware/software/Networking/ML teams to align simulation with real workloads and constraints.

• Create reproducible benchmarking methodologies across models, system configurations, and compare against real system measurements to prove validity.

• Communicate findings through technical reports and design recommendations.

Qualifications

Required:

• Master’s, or PhD in Computer Science, Electrical Engineering, Computer Engineering, or a related field.

• Strong experience in ML systems, distributed systems, performance engineering, computer architecture, or simulation.

• Understanding of systems used for machine learning training and inference.

• Experience analyzing compute, communication, and memory behavior in large-scale ML systems.

• Hands-on experience with performance benchmarking, profiling, and measurement of ML systems.

• Experience with distributed training concepts such as data parallelism, tensor/model parallelism, pipeline parallelism, collectives, and synchronization overheads.

• Proficiency in one of the following Python, C++, or Rust.

• Strong analytical skills and the ability to connect simulation results to real system behavior.

Preferred:

• Experience with system performance modelling, network simulation, or architecture evaluation tools. - this background is ideal

• Familiarity with accelerator-based systems such as GPUs, TPUs, or custom ML hardware.

• Experience with PyTorch, JAX, TensorFlow, NCCL, XLA, CUDA, or similar tools.

• Knowledge of interconnect and networking technologies such as InfiniBand, Ethernet/RDMA, NVLink, PCIe, or equivalent.

• Experience evaluating both training throughput and inference latency/serving efficiency.

• Background in workload characterization, trace-driven simulation, or model calibration.

• Ability to work across hardware and software boundaries in a cross-functional environment.
  
  

What Success Looks Like:

• Build simulation models that accurately predict performance trends and inform architectural decisions.

• Identify compute and communication bottlenecks in ML training and inference systems.

• Correlate simulation outputs with real-world benchmark data.

• Improve system efficiency, scalability, and cost effectiveness through data-driven insights.