Lead AI Platform

5 months ago

Cairo, EgyptStaff+

Responsibilities

Translate AI/ML workloads into optimized infrastructure and deployment strategies.
Optimize model latency, throughput, memory utilization, and GPU performance across environments.
Design and implement training and inference pipelines using TensorRT, Triton, and NIM.
Convert and optimize models across PyTorch, ONNX, and TensorRT.
Analyze compute, memory, and network bottlenecks using GPU and system profiling tools.
Improve GPU utilization and scheduling efficiency across clusters.
Design scalable distributed training and inference architectures for multi-GPU and multi-node systems.
Work with customers to define AI infrastructure strategies and deployment models.
Support production deployments, including monitoring, rollback, and performance validation.
Conduct applied research on model efficiency and infrastructure utilization.
Mentor team members on AI infrastructure, optimization, and GPU systems.
Use experiment tracking tools to compare parameters, metrics, and artifacts.
Investigate post-deployment model degradation, including concept drift, data pipeline changes, and traffic pattern shifts.
Perform root cause analysis for ML systems by isolating variables and reproducing issues.

8+ years of experience in AI systems and 8+ years of experience in ML systems, HPC, and AI infrastructure.
Strong proficiency in Python.
Strong experience with GPU-based AI workloads and performance optimization.
Deep understanding of quantization, pruning, batching, and other model optimization techniques.
Hands-on experience with PyTorch, ONNX, ONNX Runtime, TensorRT, TensorRT-LLM, and Triton Inference Server.
Knowledge of CUDA, cuDNN, and GPU architecture fundamentals.
Experience with distributed multi-GPU and multi-node systems.
Familiarity with NCCL, NVLink, InfiniBand, Kubernetes, or Slurm.
Experience deploying AI models into production environments.
Experience analyzing system bottlenecks and using profiling tools such as Nsight and TensorRT profiler.
Knowledge of GPU workload cost optimization strategies.
Experience with MLflow, W&B, or Neptune for experiment tracking.
Nice to have experience with NVIDIA NIM, the NGC ecosystem, Megatron-LM, NeMo, large-scale LLM training or inference, and LLM optimization techniques such as KV cache and batching strategies.
Nice to have familiarity with MLOps practices, CI/CD for AI systems, customer-facing architecture or consulting, and hybrid cloud or on-premises HPC environments.

201-500 employees