Engineering Manager, Inference Benchmarking — AI Perf

at Nvidia

USD 224,000-356,500 per year

SENIOR

✅ On-site

Tech Stack
Tag name is followed by "@" symbol and proficiency level value. About proficiency levels:

1-2 — basic awareness. Minimal hands-on experience, and a rudimentary understanding of the technology's purpose;

3-6 — daily use. Comfortable and regular usage, capable of handling common tasks and challenges related to the technology;

7-9 — you are an expert, you can teach others, you know all the pitfalls and tricks;

10 — exceptional knowledge, comprehensive understanding, and adeptness in all aspects of the technology, including advanced problem-solving. Think twice before claiming or demanding such level.

Distributed Systems @ 4 GPU @ 4 Helm @ 4 Kubernetes @ 4 LLM @ 7 Leadership @ 6 Machine Learning Mentoring Microservices Observability @ 4 Prometheus SGLang @ 4 vLLM @ 4

Driving the technical roadmap for AIPerf's core infrastructure: load generation, ZMQ-based microservices, GPU telemetry (DCGM/PyNVML, Prometheus metrics, statistical confidence intervals, and Kubernetes-native deployment.
Taking ownership for the accuracy and statistical soundness of benchmark results that engineering groups throughout the industry depend on to inform production infrastructure decisions.
Advising upstream engine integrations involving vLLM, TRT-LLM, and SGLang in partnership with NVIDIA's Dynamo and NIM teams to maintain AIPerf's relevance across emerging hardware, workload categories, and inference configurations.
Hiring, mentoring, and growing a team of senior engineers operating in a high-velocity open-source environment with active external contributors worldwide.

Bachelor's degree in Computer Science, Electrical Engineering, or related field, or equivalent experience.
8+ overall years of software engineering experience building performance-critical infrastructure, ML tooling, or distributed systems.
3+ years of engineering leadership experience as a tech lead, TLM, or engineering manager.
Deep understanding of LLM inference mechanics — TTFT, ITL, KV caching, Prefill/Decode, speculative decoding — and the ability to reason about measurement correctness and reproducibility.
Proven track record of collaborating across multi-functional groups and delivering production-quality output in high-velocity, high-external-visibility environments.

Extensive experience with vLLM, TRT-LLM or SGLang internals along with contributions to their upstream projects.
Experience building Kubernetes-native infrastructure including operators, Helm charts, and GPU observability tooling (DCGM, dcgm-exporter, PyNVML).
Background in competitive benchmarking frameworks such as MLPerf or equivalent industry-standard evaluation systems.
History leading or making meaningful contributions to active open-source projects with external communities.