Senior Software Engineer – TensorRT Edge-LLM

NVIDIA • Austin, TX

1d • Hybrid

About The Position

Are you passionate about pushing the limits of real-time large language model inference? Join NVIDIA’s TensorRT Edge-LLM team and help shape the next generation of edge AI for automotive and robotics. We build the software stack that enables Large Language, Vision-Language, and Multimodal (LLM/VLM/VLA) models to run efficiently on embedded and edge platforms — delivering cutting-edge generative AI experiences directly on-device. What you’ll be doing: Develop and evolve a state-of-the-art inference framework in modern C++ that extends TensorRT with autoregressive model serving capabilities, including speculative decoding, LoRA, MoE, and KV cache management. Design and implement compiler and runtime optimizations tailored for transformer-based models running on constrained, real-time platforms. Collaborate with teams across CUDA, kernel libraries, compilers, and robotics to deliver high-performance, production-ready solutions. Contribute to CUDA kernel and operator development for critical transformer components such as attention, GEMM, and MoE. Benchmark, profile, and optimize inference performance across diverse embedded and automotive environments. Stay ahead of the rapidly evolving LLM/VLM ecosystem and bring emerging techniques into product-grade software.

Requirements

• BS, MS, PhD, or equivalent experience in Computer Science, Electrical/Computer Engineering, or a closely related field.
• 4+ years of relevant software development experience.
• Deep understanding of transformer models and inference optimization techniques (e.g., quantization, tensor parallelism, or memory-efficient scheduling).
• Proficient programming ability with modern C++ (C++11/14/17 and beyond).
• Familiarity with popular LLM frameworks and libraries such as TensorRT, TensorRT-LLM, vLLM, SGLang, MLC-LLM, or FlashInfer.
• A track record of strong software design, execution, and collaboration across fields.

Nice To Haves

• Demonstrated development experience or open-source contributions to LLM inference frameworks and libraries, such as SGLang, vLLM, or FlashInfer.
• Proficiency with CUDA, including efficient kernel development, performance profiling, and GPU architecture fundamentals.
• Prior work on autoregressive LLM serving systems, including speculative decoding or KV cache management.
• Familiarity with compiler infrastructure for large language model inference.
• Exposure to robotics or embedded AI pipelines, including optimizing for low-latency, resource-constrained systems.

Responsibilities

• Develop and evolve a state-of-the-art inference framework in modern C++ that extends TensorRT with autoregressive model serving capabilities, including speculative decoding, LoRA, MoE, and KV cache management.
• Design and implement compiler and runtime optimizations tailored for transformer-based models running on constrained, real-time platforms.
• Collaborate with teams across CUDA, kernel libraries, compilers, and robotics to deliver high-performance, production-ready solutions.
• Contribute to CUDA kernel and operator development for critical transformer components such as attention, GEMM, and MoE.
• Benchmark, profile, and optimize inference performance across diverse embedded and automotive environments.
• Stay ahead of the rapidly evolving LLM/VLM ecosystem and bring emerging techniques into product-grade software.

Benefits

• Your base salary will be determined based on your location, experience, and the pay of employees in similar positions.
• You will also be eligible for equity and benefits.

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level