Senior AI Engineer

Immigration sponsorship is not available for this position

Responsibilities:

• Develop the Machine Learning Platform management system.

• Design and implement intuitive user interfaces and APls for seamless interaction with the platform.

• Ensure robust access control and security measures for the Machine Learning Platform.

• Regularly evaluate and enhance platform performance, scalability, and reliability. Integrate tools for data versioning, experiment tracking, and workflow orchestration.

• Build the toolchains, service, pipeline for model development workflow, and model serving architecture.

• Create automated pipelines for data preprocessing, feature engineering, and dataset versioning.

• Develop Cl/CD pipelines for deploying models into production environments with minimal downtime.

• Enable support for distributed model training and hyperparameter optimization.

• Incorporate A/B testing frameworks for evaluating multiple model deployments.

• Collaborate with data scientists and engineers to streamline the model development lifecycle.

• Prioritize various metrics for model training and inferencing monitoring. Implement logging and monitoring tools to track model performance, resource utilization, and throughput.

• Develop dashboards to visualize key metrics such as latency, accuracy, and drift detection in realtime.

• Establish alerting mechanisms to detect and respond to anomalies or performance degradation.

• Continuously refine metric prioritization based on stakeholder feedback and evolving business goals.

• Develop and maintaining the high-performance LLM training GPU infrastructure and cluster.

• Optimize GPU utilization for large-scale training workloads, ensuring minimal resource wastage.

• Implement fault-tolerant and distributed training strategies for handling large language models (LLMs).

• Evaluate and integrate emerging hardware technologies, such as TPUs, into the training infrastructure.

• Regularly update cluster configurations to support new frameworks and model architectures.

• Manage scheduling and resource allocation for multi-tenant GPU clusters.

• Understand the auto scale for inference service and multi-models for dynamical loading.

• Design systems that dynamically allocate resources based on real-time demand for inference services.

• Develop mechanisms for loading and unloading models in memory to optimize latency and resource usage.

• Implement strategies for caching frequently used models to improve inference performance.

• Experiment with serverless architectures to further enhance scalability and cost efficiency.

• Ensure compatibility with edge devices and deploy lightweight models for edge inference.

• Support, troubleshoot, and resolve any issues during the training and inferencing.

• Create detailed runbooks for common troubleshooting scenarios to reduce resolution times.

• Perform root cause analysis for failures and implement long-term fixes to prevent recurrence.

• Collaborate with DevOps and IT teams to ensure the stability of underlying infrastructure.

• Develop self-healing systems that can automatically recover from common training or inference issues.

• Provide technical support and guidance to data scientists and engineers working on the platform.

What we're looking for:

Requires a Bachelor's degree in Communications Engineering, Artificial Intelligence, Software Engineering, a related field, or a foreign degree equivalent. Must have 2 years of experience in job offered or related occupation. Must have 2 years of experience in:

• Designing, Implementing, or optimizing large-scale distributed training systems using technologies like Horovod, DeepSpeed, PyTorch Distributed, or Ray;

• Tensor/model parallelism and pipeline parallelism;

• Utilizing cloud-native or on-prem infrastructure (Kubernetes, Docker, Slurm) to support scalable, fault-tolerant, and resource-efficient AI workloads across multi-node GPU clusters;

• Using Performance Profiling and Optimization to diagnose and improve end-to-end training performance by optimizing data pipelines (e.g., DALI, tf.data), minimizing communication overhead (e.g., NCCL, gRPC), and tuning hardware-specific kernels (e.g., CUDA, Triton);

• Systems Programming and Automation in systems-level programming with Python, Bash, and C++ or Go;

• Automating deployment and orchestration of AI workloads and monitoring using Prometheus, Grafana, Weights & Biases.

• Telecommuting work arrangement permitted one day a week. Four days in office required. Position does not require domestic or international travel

Zoom Communications, Inc.
#LI-DNI
#Ind0
 

Salary Range or On Target Earnings:

Min