Principal AI Product Engineer

Houston, New York, San Francisco, Seattle

About Nscale

Nscale is taking on the hyperscalers by building a vertically integrated GenAI cloud platform. We own the data centers, software, and applications that power today's AI stack using sustainable technology solutions. We thrive on a culture of relentless innovation, ownership, and accountability, where every team member takes pride in their work and drives it with excellence and urgency. As a Nscaler, you'll build trust through openness and transparency, where everyone is inspired to do their best work. Collaboration is key, and we work together swiftly and respectfully, embracing adaptability and resilience in all we do.

About the Position

Nscale is looking for a Principal AI Engineer (Specialised) to lead the inference and post-training pillar of our AI systems engineering organization. You’ll define the multi-year technical roadmap for how models are served, evaluated, and post-trained on Nscale’s GPU cloud, across dedicated and serverless inference and bring-your-own-model deployments. You’ll lead the most consequential architectural programs in that space and set the engineering standards that 20–50+ engineers build to.

As a Principal engineer, you are one of the deepest technical authorities in the company on AI systems. Your decisions set the cost, latency, and reliability at which Nscale serves tokens and runs post-training workloads, and those numbers have to compete with the world’s leading AI infrastructure providers. The problems span the full stack: kernel efficiency on state-of-the-art GPU systems, fleet-level KV cache and serving architecture, the evals that prove model quality, and RL loops where inference and training share hardware. You frame the solutions the organization executes against, including the API contracts customers see.

Responsibilities

• Define and own the multi-year technical roadmap for Nscale’s inference, evals, and post-training platform, and translate it into architecture that multiple teams can execute against
• Lead company-scale architectural initiatives in the pillar, such as next-generation serving (disaggregated prefill/decode, KV cache orchestration across GPU, host, and storage tiers, speculative decoding, multi-tenant scheduling), GPU kernel and model efficiency work (custom kernels, FP8/NVFP4/INT8/4 quantization, sparsity, distillation, MoE serving), evals and benchmarking frameworks, and post-training and RL infrastructure
• Establish engineering standards adopted across all AI teams: API design and compatibility guarantees, benchmarking and evals methodology, training stability norms, and performance testing practices
• Own the framework by which cost, latency, throughput, and model quality trade-offs are made and measured across the pillar
• Identify long-horizon systemic risks early (serving engine and framework bets, accelerator support, capability gaps) and resolve them before they block the organization
• Align AI engineering, research, product, and infrastructure leadership on multi-team technical strategy; frame technical trade-offs in product and commercial terms
• Mentor and develop Staff and Senior AI Engineers, and grow the next generation of inference technical leaders at Nscale
• Represent Nscale’s technical approach externally: open-source leadership in the frameworks we depend on, publications, conference talks, and partnerships with GPU vendors and AI labs

Requirements

• 10–15 years of engineering experience, with a clear track record of pillar-level impact on production AI systems
• 4+ years of hands-on work with LLMs in inference, GPU performance, evals, or post-training and RL, in production or research
• Demonstrated ability to define multi-year technical strategy for complex, multi-team AI systems organizations
• World-class depth in production LLM inference, GPU performance, evals, and/or post-training and RL infrastructure, with strong working knowledge across the rest
• Demonstrated ownership of the architecture of a large-scale production inference or training platform
• Proven ability to create architectural frameworks and engineering standards adopted across large engineering organizations
• Deep understanding of the hardware/software boundary for AI accelerators: CUDA or ROCm, memory bandwidth and interconnect constraints, and distributed compute paradigms
• Strong history of growing technical leaders (Staff and above) and multiplying technical capability across teams
• External recognition in the AI systems community through research, open source, or industry contribution

Preferred

• Prior experience at a top-tier AI lab or major hyperscaler AI infrastructure team
• Maintainer or core contributor to a foundational inference, kernel, or RL framework (vLLM, SGLang, TensorRT-LLM, LMCache, FlashInfer, Triton, verl, OpenRLHF, TRL, DeepSpeed, Megatron-LM, etc.)
• Hands-on depth in RL for LLMs (DPO/GRPO-style methods, reward modelling, multi-turn and tool-use RL) and the interaction between inference and training infrastructure
• Experience defining developer API platforms adopted at scale by external developers
• Deep experience with control plane / data plane architecture and cell-based deployment patterns in large-scale inference infrastructure
• Published work in AI systems: MLSys, NeurIPS Systems Track, OSDI, EuroSys, SC, or equivalent
• Experience with hardware-software co-design: custom accelerator kernels (CUDA, Triton), compiler-level optimization, AI hardware roadmap engagement
• Experience defining pricing, SLO, and capacity models for a commercial inference product

We Offer

• Competitive salary range: $290,000 - $520,000 USD
• Comprehensive benefits package including medical, dental, vision, flexible paid time off, parental leave, and retirement plan participation
• Opportunity to work in a dynamic and innovative environment

For information on how Nscale handles candidate personal data, please see our Employee & Candidate Privacy Notice: Here. Nscale does not accept unsolicited candidate submissions from recruitment agencies.

Apply for this Job

Apply here