Senior NVIDIA AI Platform Architect

Kyndryl Inc.

Jakarta Pusat

On-site

IDR 800,000,000 - 1,500,000,000

Full time

2 days ago

Be an early applicant

Application generator

Stand out for this role — generate a tailored resume and cover letter in about a minute.

Get past ATS filters

Job summary

Kyndryl is seeking a senior engineer to own production-grade NVIDIA AI platforms in Jakarta. You will tackle cross-layer issues across Linux, CUDA, Docker/Kubernetes, and GPU health, shaping standards for reliable training and inference environments.

You will lead RCA, build runbooks, mentor engineers, and drive platform upgrades and performance investigations across multi-GPU systems.

Qualifications

Strong Python and Bash automation, observability engineering and deep platform troubleshooting.
Strong NCCL, multi-GPU or multi-node performance troubleshooting, platform lifecycle upgrades, canary and rollback experience.
Strong PyTorch expertise plus one or more of vLLM, llama.cpp, TensorRT, Triton, NIM, NGC or NeMo.
Expert Linux, NVIDIA driver, CUDA, container runtime and Kubernetes knowledge, with deep hands-on experience in NVIDIA GPU Operator, DCGM and multi-GPU scheduling and health.

Responsibilities

Lead Root Cause Analysis (RCA), automate recurring operations, develop Level 1 (L1) and Level 2 (L2) runbooks, mentor engineers and act as technical authority during Priority 0 / Priority 1 (P0/P1) incidents.
Define platform compatibility, upgrade, canary and rollback standards, and design GPU observability and remediation thresholds using DCGM, Prometheus, Grafana or equivalent tooling.
Architect and optimize local or cloud-based AI training and inference platforms using PyTorch and NVIDIA technologies such as TensorRT, Triton, NVIDIA Inference Microservices (NIM), NVIDIA GPU Cloud (NGC) and NVIDIA NeMo.
Lead NVIDIA Collective Communications Library (NCCL) and distributed multi-GPU or multi-node performance investigations across compute, network/fabric and storage.
Lead complex diagnosis across Linux, NVIDIA drivers, CUDA, container runtime, Kubernetes or Red Hat OpenShift, NVIDIA GPU Operator, Data Center GPU Manager (DCGM) and AI workloads, including recurring NVIDIA driver error identifiers (XIDs) and Multi-Instance GPU (MIG) issues.