ML and Agentic Systems Engineer

NVIDIA•Us, CA

21h

About The Position

At NVIDIA, we’re not just building the future, we’re generating it! Our Cosmos team is pushing the boundaries of multimodal AI, simulation, and world models. As we enter the next phase, we are building agentic systems that can reason about, build, evaluate, and improve AI systems themselves. We are building systems where AI doesn’t just run models but helps build them. This role is about creating the meta-layer of modern ML: the agents, tooling, pipelines, and feedback loops that make model development faster, smarter, and increasingly automated. Rather than focusing on inventing individual model architectures, you will build the systems that help models and teams improve continuously. We are looking for exceptional engineers who are passionate about the idea of AI-native software engineering: systems where agents can work with code, data, experiments, and evaluations to accelerate how machine learning gets done.

Requirements

Significant experience building machine learning systems and software platforms, not only models.
Expert-level Python skills, with strong judgment around modularity, abstraction boundaries, and long-term code health.
Deep familiarity with PyTorch, including the ability to debug, adapt, and extend model behavior within larger software systems.
Experience building pipelines, evaluation systems, developer tooling, or workflow automation for ML at meaningful scale.
Strong software engineering fundamentals, including system design, testing, packaging, debugging, and collaborative codebase evolution.
Strong agency in LLM-based systems, such as tool use, planning, multi-step workflows, code agents, or automation over data and experiments.
Comfort operating in fast-moving environments where ambiguous ideas must be turned into useful systems quickly.
BS, MS, or equivalent experience in Computer Science, Engineering, or a related field.
12+ years of relevant software development experience

Nice To Haves

You have built agent-based systems that do real work: coding, evaluation, data generation, triage, experimentation, or orchestration.
You have contributed to impactful open-source ML, Python, or developer tooling.
Background with context compression and agent memory techniques
Familiarity with agent safety and agent identity (AuthN, AuthZ, IAM)
You bring a high bar for software craftsmanship, but know how to apply it in research-adjacent environments without slowing innovation down.

Responsibilities

Design and implement agentic workflows across the ML lifecycle, including data generation and curation, evaluation, debugging, training orchestration, and iteration.
Build AI-native systems in which models and agents can interact with codebases, tools, experiments, and environments to improve developer and researcher productivity.
Create self-improving loops where agents help generate data, surface failures, evaluate outputs, and drive better decisions across the system.
Own and evolve large-scale Python and PyTorch codebases, turning fast-moving ideas into robust, modular, reusable software.
Design and scale evaluation platforms that combine automated metrics, human feedback, and agent-driven analysis.
Build and maintain multimodal ML pipelines spanning data processing, experimentation, benchmarking, and deployment.
Integrate open-source and internal components into unified systems that enable rapid experimentation and reliable iteration.
Raise the bar on engineering excellence across the team through strong practices in testing, reproducibility, packaging, code health, and maintainability.