Architecture · Beijing, China

Full-time

Deep Learning Solution Architect - Agentic Performance

NVIDIA · Beijing, China · Posted Aug 27, 2026

This posting comes from NVIDIA. Confirm the details on their site before applying.

Overview

NVIDIA are seeking dynamic Solution Architects with specialized expertise in training Large Language Models (LLMs), implementing RAG workflows, and agentic inference. You will leverage the full NVIDIA software & hardware ecosystem to design, optimize, and deliver production-grade generative AI solutions for enterprise customers. With competitive salaries and a generous benefits package, we are widely considered to be one of the world’s most desirable employers!

Responsibilities

  • Architect end-to-end solutions focused on LLM pretraining, fine-tuning, high-performance inference, RAG workflows, and agentic inference orchestration using NVIDIA’s hardware and software platforms.
  • Collaborate with customers to understand their LLM-related business challenges and design tailored solutions aligned with the NVIDIA ecosystem.
  • Lead LLM training, distributed optimization, and performance tuning to achieve optimal throughput, latency, and memory efficiency.
  • Design and integrate RAG workflows and agentic inference pipelines into customer systems; provide technical guidance on best practices.
  • Collaborate with NVIDIA engineering teams to provide feedback and support pre-sales technical activities (workshops, demos).
  • Master’s / Ph.D. in Computer Science, Artificial Intelligence, or equivalent experience.
  • 4+ years hands-on experience in AI, focusing on open-source LLM training, fine-tuning, and production inference optimization.
  • Deep understanding of mainstream LLM architectures and proficiency in LLM customization via PyTorch, Hugging Face Transformers.
  • Solid knowledge of GPU computing, cluster architecture, and distributed parallel training/inference for LLMs.
  • Competency in agentic inference design and using AI agents to solve business challenges.
  • Strong communication skills, able to articulate complex technical concepts to technical and non-technical stakeholders.

Preferred qualifications

  • Hands-on experience with NVIDIA’s generative AI ecosystem (TRT-LLM, Megatron-LM, NVIDIA NeMo).
  • Advanced skills in LLM optimization (quantization, KV Cache tuning, memory footprint reduction).
  • Experience with Docker, Kubernetes for containerized LLM and agent workflow deployment on-prem.
  • In-depth knowledge of multi-GPU parallelism and large-scale GPU cluster management.

Great fits start here.

Whether you're looking for your next role or your next team member, give people and possibilities a place to meet.