Asks for

The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”
The ad says“Enough ML fluency to work productively with AI researchers: you understand training loops, reward signals, and inference-time behavior well enough to debug them, even without designing the algorithms yourself.”
The ad says“Enough ML fluency to work productively with AI researchers: you understand training loops, reward signals, and inference-time behavior well enough to debug them, even without designing the algorithms yourself.”
The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”
The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”
The ad says“Enough ML fluency to work productively with AI researchers: you understand training loops, reward signals, and inference-time behavior well enough to debug them, even without designing the algorithms yourself.”
The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”
The ad says“Strong software engineering fundamentals. You can build a service that other engineers and researchers depend on every day, not a script that worked once.”
The ad says“Enough ML fluency to work productively with AI researchers: you understand training loops, reward signals, and inference-time behavior well enough to debug them, even without designing the algorithms yourself.”
The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”
The ad says“Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).”

Overview

Physical Superintelligence is a startup with roots at Google, NVIDIA, Harvard, Meta, MIT, Oxford, Johns Hopkins, Cambridge, and the Perimeter Institute building AI systems to discover new physics at scale.

Our mission is to discover and commercialize transformative physics breakthroughs at scale with artificial superintelligence, safely, verifiably, and for broad public benefit.

The last century's golden age of physics gave us transistors, lasers, and nuclear energy. We believe artificial superintelligence will unlock the next one. We're creating the infrastructure to industrialize scientific discovery and usher in this new era.

We have one product: new physics, at scale.

Role

We are seeking a Member of Technical Staff, ML Engineer to build and run the training and inference systems that turn Core AI's research into things that work at scale, and that the rest of Engineering can build on.

Responsibilities

Own the training and inference infrastructure that Core AI depends on: distributed training jobs, GPU scheduling, and model-serving systems (vLLM, SGLang, or comparable) for both proprietary models and self-hosted inference.
Build the tools and abstractions AI researchers use to launch training runs, iterate on inference providers, and route workloads across models, so a researcher's time goes into the science instead of the plumbing.
Partner with Engineering on the shared platform: capacity planning, observability, and reliability for GPU and inference infrastructure, so training and serving hold up to the same production bar as everything else we ship.
Debug and harden the training and inference stack under real load. Egress failures, stalled retries, and routing edge cases are your problem to close, not someone else's ticket.
Stay hands-on. You write the code, not just the design doc, and you are the first call when a training job stalls or an inference path breaks.

What We're Looking For

Three or more years building and operating ML training or inference infrastructure in production, at a company that trains or serves models at meaningful scale.
Hands-on experience with distributed training (multi-GPU or multi-node, using PyTorch, Ray, or comparable) and model-serving systems (vLLM, SGLang, Triton, or comparable).
Strong software engineering fundamentals. You can build a service that other engineers and researchers depend on every day, not a script that worked once.
Enough ML fluency to work productively with AI researchers: you understand training loops, reward signals, and inference-time behavior well enough to debug them, even without designing the algorithms yourself.

Nice to Have

Experience building internal platform tools such as training-as-a-service APIs, inference gateways, or job schedulers.
Background in GPU infrastructure, CUDA, or performance engineering for ML workloads.
Experience with cloud infrastructure (GCP, AWS) and infrastructure as code (Terraform or comparable).
Prior work embedded alongside a research team, turning research code into production systems.

How We Work

We hold a high technical bar and give people full ownership of their work, from spec to ship to on-call. We write contracts before logic, test against real systems instead of mocks, and favor simple designs that ship over clever ones that do not. Our development process is AI-native: we work with agentic coding tools daily, write specs that are legible to humans and agents alike, and lead with leverage.

Location and Compensation

This role is based in Boston. We will consider remote candidates on a case-by-case basis. We offer competitive compensation including salary, benefits, and meaningful early-stage equity. We evaluate on technical breadth, systems thinking, ML infrastructure depth, and shipping velocity. We are an equal opportunity employer and value diverse perspectives in building platforms for AI-driven discovery.