Senior Technical Program Manager - LLM Safety

NVIDIA•Santa Clara, CA

10d

About The Position

NVIDIA is seeking a Senior Technical Program Manager to lead safety planning and execution for Nemotron models, with a focus on LLM and agentic safety. The role leads cross-functional evaluation programs, tracks evaluation results and analysis, and turns findings into prioritized mitigation plans with accountable owners, committed dates, and measurable closure criteria. The position establishes evaluation coverage across multi-turn, multimodal, multilingual, long-context, and reasoning use cases, and creates governance for reviewing findings, assessing severity and release impact, and escalating unresolved risks. It ensures safety evidence, approvals, exceptions, and risk-acceptance decisions are documented before release. The role also identifies program gaps and dependencies, drives decisions and corrective actions, and builds dashboards and executive reports covering evaluation coverage, critical findings, mitigation progress, residual risk, and release readiness.

Requirements

Bachelor’s degree or equivalent experience in computer science, engineering, data science, or a related technical field.
10+ years of experience in technical program management, engineering, product development, technical operations, or a similar area.
Strong understanding of LLM architecture, frameworks such as OpenAI, Anthropic, and Hugging Face, and model evaluation.
Familiarity with LLM development, post-training, inference, tool calling, evaluation datasets, and model-release lifecycles.
Experience directing programs related to AI/ML development, model evaluation, agentic safety, security, content safety, hallucinations, and/or production releases.
Familiarity with AI safety risks such as hallucinations, timely injection attacks, unsafe tool use, data poisoning, model manipulation, and unintended agent behavior.
Experience supporting model-safety evaluations, red teaming, adversarial testing, security assessments, or responsible-AI programs.
Ability to interpret technical evaluation findings and communicate their product, schedule, and release implications to leadership.

Nice To Haves

Experience with LLM agentic safety, security, red teaming, hallucinations, adversarial assessment, or responsible AI.
Experience defining evaluation datasets, rubrics, benchmarks, and release gates.
Ability to use evaluation results and data to make clear release recommendations and communicate residual risk to leadership.
Experience crafting or operationalizing multi-turn, hallucination, adversarial, or agentic-safety evaluations.

Responsibilities

Lead cross-functional planning and execution for identifying, selecting, and running evaluations, especially agentic-safety evaluations involving multi-turn tool calling, unsafe tool use, unintended actions, excessive autonomy, and multi-step failures.
Track evaluation execution and results, analysis, and resulting mitigation and research plans for improvement.
Translate evaluation findings into prioritized mitigation plans with accountable owners, committed dates, and measurable closure criteria.
Lead end-to-end safety planning and execution for Nemotron models, including scope, dependencies, risks, resources, and release-readiness criteria.
Translate technical safety priorities into executable programs covering LLM security, frontier risks, agentic safety, and hallucinations.
Establish and track evaluations across multi-turn, multimodal, multilingual, long-context, and reasoning models, covering relevant use cases, domains, languages, modalities, and model releases.
Establish governance for reviewing safety findings, assigning severity, determining release impact, and escalating unresolved risks.
Ensure safety evidence, approvals, exceptions, and risk-acceptance decisions are documented before release.
Identify program gaps and critical dependencies early, and drive decisions, escalations, recovery plans, and corrective actions.
Build and maintain dashboards and executive reporting for evaluation coverage, critical findings, mitigation progress, residual risk, and release readiness.