Principal Product Manager - AI/ML Training at Scale

Amazon Web Services (AWS); Seattle (WA); On-site; USD 181,000 - 245,000; Full time

Job summary

Annapurna Labs (U.S.) Inc. in Seattle, WA seeks a Principal Technical Product Manager to define and drive training software strategy for Trainium. You will shape distributed training libraries, post-training workflows, and optimization to empower researchers and operators to scale frontier models.

You will collaborate with engineering, science, and BD teams to align roadmaps, drive developer experience, and push ahead of evolving training paradigms across Neuron tooling.

Qualifications

• 7+ years as a Technical Product Manager or similar role.
• Experience with large-scale model training workflows and distributed training.
• Familiarity with major AI/ML frameworks such as JAX or PyTorch.
• Strong ability to drive product strategy and cross-organizational alignment.
• Excellent written and verbal communication, including executive-level interactions.

Responsibilities

• Define and execute training product strategy and roadmap for Trainium.
• Lead post-training workflows including RLHF, DPO, fine-tuning at scale.
• Engage customers and internal teams to translate needs into requirements and metrics.
• Own training ecosystem tools integration and developer experience improvements.

Skills

Technical Product Manager experience

Large-scale model training workflows

Distributed training concepts

Executive-level communication

JAX or PyTorch familiarity

Education

Bachelor's degree or equivalent

Tools

PyTorch

JAX

TorchTitan

TRL

VeRL

MaxText

AXLearn

Job description

Annapurna Labs (U.S.) Inc. in Seattle, WA seeks a Principal Technical Product Manager to define and drive training software strategy for Trainium. You will shape distributed training libraries, post-training workflows, and optimization to empower researchers and operators to scale frontier models.

You will collaborate with engineering, science, and BD teams to align roadmaps, drive developer experience, and push ahead of evolving training paradigms across Neuron tooling.