Deep Learning Engineer, Accuracy Evaluation

NVIDIA Corporation; United States; Remote; USD 180,000 - 260,000; Full time

Job summary

NVIDIA Corporation ищет старших инженеров для разработки методик оценки передовых моделей их архитектуры, включая LLM, RAG, агентов и визуальные модели. Ваша работа будет охватывать создание точных сигналов для выпуска и направления исследований, а также работу на мощных GPU‑кластерах по всему Enterprise.

Требуется 6+ лет опыта в области оценки больших языковых моделей и мультимодальных систем, опыт построения инфраструктуры CI/CD и анализа модернизированных данных.

Qualifications

• Учитывать требования: наличие степени BS/MS/PhD в соответствующей области.

Responsibilities

• Разрабатывать среды оценки для моделей frontier: рассуждения, мультимодальные и агентные системы.
• Исследовать новые методики оценки для новых семейств моделей и доменов возможностей.
• Разрабатывать инфраструктуру оценки: пайплайны, регрессивные CI, статистический анализ.
• Согласовывать сигналы оценки с решениями по релизу и направлению обучения.

Skills

Оценка LLM

Агентные системы

Статистический дизайн

Экспериментальный дизайн

Анализ Benchmark

Четкая коммуникация

Education

BS/MS/PhD в CS/ML/Statistics

Tools

MLflow

W&B

SWE-bench

GAIA

WebArena-style

Job description

Описание: We are seeking senior engineers to pioneer new methodologies for accurately assessing the performance and capabilities of ground-breaking deep learning models, including LLMs, RAG, agents, and vision models. You will collaborate across the organization to bring the latest flagship models from our community and partners to life. This role offers an outstanding opportunity to craft the future of AI at a fast-growing company at the frontward of the AI revolution. Join our team of world‑class software engineers and partners to deliver the most advanced models with lightning‑fast inference. You'll work on the most powerful, enterprise‑grade GPU clusters capable of hundreds of PetaFLOPS and gain early access to unreleased hardware, making a direct impact on NVIDIA's roadmap and the broader AI landscape!

Задачи:

• Design and build decision‑grade evaluation environments for NVIDIA's frontier models spanning reasoning, multimodal, long‑context, and agentic systems, producing auditable accuracy signals that gate every major model release.
• Research and develop novel evaluation methodologies for emerging model families and capability domains (low‑precision numerics, multi‑turn agentic tasks, code generation) where established benchmarks don't yet exist or don't generalize.
• Build and operate the evaluation infrastructure and pipelines including benchmark environments, regression CI systems, and statistical analysis tooling, used by model, product, and applied research teams across NVIDIA.
• Partner with model research, training, and customer teams to translate evaluation signals into concrete decisions: release go/no‑go, training iteration direction, and competitive positioning against external frontier models.

What we need to see: BS, MS, or PhD in Computer Science, Machine Learning, Statistics, or a related field 6+ years of hands‑on experience with LLMs, designing and running evaluations for large language models or multimodal AI systems, including experience with agentic, multi‑turn, or reasoning‑heavy settings. Strong statistical foundations: experimental design, significance testing, regression analysis, and the ability to distinguish signal from noise in benchmark results at scale. Proven experience building evaluation infrastructure (pipelines, benchmark harnesses, reproducible CI systems) not just consuming existing benchmarks. Clear, precise communicator who can translate quantitative evaluation results into decisions for researchers, product teams, and senior leadership.

Требования:

• Deep familiarity with open‑source evaluation frameworks Experience designing evaluations for agentic systems: tool use, multi‑turn reasoning, environment‑based benchmarks (SWE‑bench, GAIA, WebArena‑style), or interactive evaluation settings.
• Track record of publishing or contributing to evaluation research, new benchmark design, methodology papers, or reproducibility analyses that shaped how the field measures model capability.
• Experience measuring model accuracy under low‑precision inference (FP8, INT4, quantization‑aware settings) and understanding how calibration and sparsity interact with benchmark results.
• Comfort running large‑scale workloads on HPC/Slurm clusters, including reproducible experiment management (MLflow, W&B) and compute cost optimization across hundreds of benchmark runs.

NVIDIA is committed to fostering a diverse work environment and proud to be an equal opportunity employer. As we highly value diversity in our current and future employees, we do not discriminate (including in our hiring and promotion practices) on the basis of race, religion, color, national origin, gender, gender expression, sexual orientation, age, marital status, veteran status, disability status or any other characteristic protected by law. Your base salary will be determined based on your location, experience, and the pay of employees in similar positions. For Poland: The base salary range is 375,000 PLN - 650,000 PLN.