Principal Product Manager, AI Frameworks

NVIDIA•Us, CA

23h•Hybrid

About The Position

At NVIDIA, we meet customers where they are on their AI journey on our GPUs - this means we build best in class frameworks in OSS and support a robust ecosystem of other OSS frameworks. NVIDIA's PyTorch Compilers team builds and upstreams to the stack that sits between PyTorch and NVIDIA hardware — spanning torch.compile, emerging compiler substrates, and the agent-native optimization infrastructure being built for the next era of accelerated computing. This role will build and direct product strategy across the full arc. It involves shipping the latest hardware features in torch.compile today. It also includes developing the canonical shared representation across NVIDIA's compiler and runtime ecosystem. Additionally, it focuses on crafting how agents will engage in deep learning performance work in the future. We are looking for someone who understands compilers and can operate at the intersection of systems architecture, framework engineering, and customer-facing product strategy working directly with engineering leadership and NVIDIA's most sophisticated external customers — including frontier model labs, inference/training and RL framework teams such as vLLM, SGLang, torchtitan, megatron-core, and hardware co-design programs. As NVIDIA Product Managers, we partner with NVIDIA leaders to define clear product strategy, and marketing team teams to build go-to-market plans. The Product Management organization at NVIDIA is a flexible, strong, and impactful group focusing on enabling deep learning across all GPU use cases and providing great products for our users. We seek an individual with a rare blend of product skills, technical depth, and passion to join our team. Does that sounds familiar? If so, we would love to hear from you!

Requirements

15+ years in technical product management, with 5 years owning a compiler, runtime, or low-level systems product at scale
BS or MS degree in Computer Science, Electrical Engineering, a related technical field, or equivalent experience.
Experience with OSS-first products and upstream contribution strategy
Track record of shipping and driving adoption for developer efficiency and performance oriented infrastructure products
Understands how inference frameworks use compiler technology — where they adopt torch.compile, where they go around it, and why
Understands how new hardware features create new compiler requirements
Can write clear, defensible strategy documents and knows how to scope an early-access release: Strong instinct for where to concentrate investment vs. spread it

Nice To Haves

Deep understanding of the PyTorch compiler stack and how to influence strategy in this ecosystem
You've worked on how agents participate in systems-level optimization workflows
Familiarity with MLIR-based compiler infrastructure and how it maps to NVIDIA hardware primitives
Can reason about inter-kernel optimization tradeoffs to define the right bar for "proven"
Comfortable reading kernel performance profiles and debug how torch.compile can help any model

Responsibilities

Own the strategy for NVIDIA's PyTorch compiler portfolio , including: torch.compile — maintain and evolve NVIDIA's upstream PyTorch path, drive HW support, resolve customer issues across dynamic shapes, kernel performance, and compile overhead
define the roadmap and go-to-market for NVIDIA's shared representation layer across frameworks, compilers, kernel libraries, and runtimes
Agent-native compiler workflows — shape the product vision for how agentic systems will participate in optimization tasks
Inter-kernel optimization features — build product requirements for capabilities like megakernels and ensure they ship with demonstrated real-world value
Lead product strategy across the roadmap: Define Now/Next/Later priorities in close partnership with engineering leads
Translate ecosystem signals (vLLM RFCs, SGLang CUDA Graph patterns, TorchTitan GraphTrainer plans, Meta's upstream priorities) into prioritized product decisions
Engage directly with customers and partners:
Represent NVIDIA at PyTorch contributor and ecosystem forums; Define success metrics and release criteria: Establish performance gates and adoption milestones
Set bar for what "proven" means for new optimizations before committing to roadmap