Thinking Machines Lab · posted · checked
Site Reliability Engineer, Post Training
C++LinuxPython$300k – $400kOn-site · New York and San FranciscoMid-level
Who can apply
On-site · New York and San Francisco
Location: New York / San Francisco
ABOUT THINKING MACHINES
The mission of Thinking Machines is to build AI that extends human will and judgment. We are training frontier models with Inkling, developing Tinker to let people make models their own, and crafting interfaces that broaden human-AI communication. We believe the future worth building is human, and we're hiring people who want to build it.
ABOUT THE ROLE
We're hiring a Site Reliability Engineer (SRE) to keep our post-training and reinforcement learning (RL) systems fast, reliable, and easy for researchers to iterate on. Think of this as a production engineering or site reliability role built around model training: you'll own the health of the training runs, clusters, and pipelines that power post-training and RL at Thinking Machines.
You'll work side by side with research teams during active model runs — debugging failures in real time, hardening infrastructure against the next class of problem, and building the tooling and automation that let researchers spend their time on the science instead of babysitting jobs. This role has real ownership: you'll be the person a research team calls when a run stalls at 2am, and the person who makes sure it doesn't happen again.
WHAT YOU’LL DO
Own the reliability, performance, and uptime of large-scale post-training and RL training jobs, from launch through completion
Partner directly with research teams during active model runs, embedding with them to unblock training and speed up iteration
Debug failures across the full stack — accelerators, networking, storage, schedulers, and training frameworks — and drive issues to root cause
Build monitoring, alerting, and automated recovery so runs self-heal or fail fast instead of silently stalling
Improve checkpointing, fault tolerance, and job scheduling so hardware failures cost minutes, not days of compute
Build internal tools that reduce toil and improve cluster utilization across post-training and RL workloads
Participate in an on-call rotation supporting production model runs
Write postmortems and turn recurring failure patterns into permanent infrastructure fixes