Site Reliability Engineer, Post Training

Thinking Machines Lab•San Francisco, CA

42d•$350,000 - $475,000•Onsite

About The Position

Thinking Machines is seeking a Site Reliability Engineer (SRE) to ensure the speed, reliability, and ease of iteration for their post-training and reinforcement learning (RL) systems. This role involves owning the health of training runs, clusters, and pipelines that support these advanced AI processes. The SRE will collaborate closely with research teams, debugging failures in real-time, strengthening infrastructure, and developing automation to allow researchers to focus on scientific discovery. This position offers significant ownership, including being the primary contact for critical issues during active model runs and implementing solutions to prevent recurrence.

Requirements

4+ years of experience as a production engineer, site reliability engineer, or infrastructure engineer operating large-scale distributed systems in production
Track record debugging complex failures across distributed systems — networking, hardware, kernel, or scheduler issues
Strong software engineering skills in Python and/or Go/C++, with the judgment to know when to script a fix versus build a system
Solid grounding in Linux systems internals and networking fundamentals
Comfortable owning production systems, including participating in on-call rotations

Nice To Haves

Experience operating GPU or TPU training clusters at scale
Familiarity with post-training and RL techniques (e.g., RLHF, PPO, DPO) and the infrastructure challenges specific to them, such as reward model serving, rollout generation, and mixed training/inference workloads
Experience with distributed training frameworks (e.g., PyTorch, Ray) and job schedulers (e.g., Slurm, Kubernetes)
Experience with high-performance networking (e.g., InfiniBand, RDMA, NCCL) and its role in distributed training performance
Experience building observability tooling purpose-built for ML training, not just general infrastructure
A track record of thriving in fast-changing, research-driven environments where priorities shift with the science

Responsibilities

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