Software Engineer, OpenShell

NVIDIA•Us, CA

About The Position

Are you ready to translate groundbreaking AI research into secure, production-grade systems? Want to shape the next generation of AI agent infrastructure? Join us! At NVIDIA OpenShell, we are building the runtime infrastructure for secure, scalable, production-grade AI agents. As an OpenShell Research Engineer, we will look to you to help bring the latest advances in agentic systems into the runtime, tools, and workflows that enterprise builders rely on. This role sits at the intersection of research, product, and engineering. We expect you to identify promising methods from academia, industry, open source, and internal NVIDIA research, understand where they matter for OpenShell, validate their impact through hands-on prototypes, benchmarks, and real agent workflows, and help us integrate the best ideas into the product.

Requirements

8+ years of professional practical experience in research engineering, software development, or a related technical field
MS/PhD in Computer Science, Physics, or a related field or equivalent experience
A strong background in turning complex research into reusable products, tools, demos, benchmarks, or production systems at scale.
Deep experience in several of the following: LLMs, agent harnesses, multimodal generative models, evaluation frameworks, synthetic data generation, post-training, inference infrastructure/optimization, adversarial ML, or agent safety/security.
Demonstrated ability to drive independent technical investigation: survey relevant work, run experiments, form a clear point of view, and communicate findings clearly.
Strong product sense and care for UX and AX: tools should be intuitive for developers and ergonomic for agents.
A focus on real-world impact: we want research to become enterprise capabilities, reference implementations, developer workflows, or product improvements.
Outstanding team orientation and comfort collaborating across research, engineering, product, design, solutions, and developer-facing teams.

Nice To Haves

Experience with secure agent runtimes, tool sandboxing, capability-based security, or enterprise policy systems.
Experience with compliance or enterprise governance requirements such as auditability, data retention, access control, SOC2, HIPAA, GDPR, or regulated deployment environments.
Experience with LLM inference infrastructure, model serving, or inference optimization using tools such as Triton, TensorRT-LLM, vLLM, SGLang, Ray, Kubernetes, or cloud GPU platforms.
Experience integrating inference backends into agentic systems, including routing across models, tool-aware context management, streaming, structured outputs, retries, monitoring, and cost/performance optimization.
Experience developing or maintaining open-source software in AI agents, LLM systems, developer tooling, ML infrastructure, model serving, or related areas.

Responsibilities

Track the cutting edge: How are agentic systems evolving? You'll follow research in tool use, planning, memory, evaluation, self-improvement, multi-agent workflows, runtime infrastructure, and agent safety/security.
Bridge research and product: Identify research ideas that can meaningfully improve OpenShell and translate them into concrete product opportunities.
Benchmark and adapt: Reproduce and test promising methods from papers, open-source projects, industry work, and internal NVIDIA research.
Build rapid prototypes: Create hands-on proof-of-concepts using OpenShell, including agent harnesses, evaluation loops, self-improving workflows, and runtime-native developer experiences.
Red-team systems: Design evaluation and red-team harnesses that measure agent reliability, usefulness, scalability, safety, security, and developer experience.
Secure the workflow: Help us design secure-by-default workflows for agents operating with tools, code, files, credentials, and enterprise systems.
Partner across teams: Collaborate closely with engineering, product, design, research, solutions, and developer-facing teams to move ideas from prototype to product.