Research Engineer - Distributed Training & RL Environment Infrastructure
ML Infrastructure Engineer - Distributed Training & SystemsSan Francisco, CA | Hybrid | Full-Time | Permanent Employment | Mid-Level | ASAP StartMaxIT Consulting - Max Corporate Group is looking for a Research Engineer to build the training and environment infrastructure of a fast-growing AI company in San Francisco. The company builds evaluation environments and post-training infrastructure for long-horizon enterprise agents, and uses that stack to train and deploy custom models for very large technology companies.
It is a small, founding-stage team out of top AI labs and technology companies, already profitable and with real enterprise revenue. This role sits under the whole company: the research and deployed teams build directly on what you own.
THE CHALLENGEAs agent workflows get longer, evaluation and training infrastructure become the bottleneck. You will own the systems that let long runs be paused, rewound and continued, and that start, watch and shut down thousands of heterogeneous environments cheaply and reliably.
Within your first 90 days you will own one of the core systems end to end: checkpoint and restore, the environment orchestrator, or the simulated companies used for training and evaluation.
KEY RESPONSIBILITIES- Build checkpoint and restore of full machine state, including memory and GPU state, across many machines.
- Build the orchestrator that starts, monitors and shuts down thousands of environments, from a single container to a multi-GPU cluster.
- Provision real cloud accounts for agents, with hundreds created and deleted on demand.
- Build realistic simulated companies with code, people, tickets, threads and documentation for agent training and evaluation.
- Be the infrastructure foundation that the research and deployed teams depend on.
WHAT WE ARE LOOKING FOR- One to five years of full-time experience (at least one year full time).
- Deep experience in at least one of: distributed systems, containers and operating-system internals, ML infrastructure or model training.
- Has trained or served models at real scale (distributed training, inference or serving), or built the environment or benchmark behind a published result.
- A publication at a top venue (such as ICLR or NeurIPS).
- Has built and owned infrastructure that other people depended on, beyond a single research script.
- Comfortable choosing the right tools: no specific language or framework is required.
- Professional English, as the role works daily with technical teams and enterprise customers.
NICE TO HAVE- Experience with checkpointing, cluster orchestration or scheduling of many heterogeneous jobs.
- Compiler, inference or serving work on large models.
- Simulation or benchmark environments (robotics sim, agent gyms, browser or operating-system environments, multi-agent benchmarks).
- Master's degree from a strong computer science program.
- Interest in joining an early-stage company for the long-term upside.
POSITION DETAILSWork arrangement: Hybrid, three to four days a week in the San Francisco office
Schedule: Full-Time, six-day work week
Employment type: Permanent employment
Seniority: Mid-Level (1 to 5 years)
Start date: ASAP
Location: San Francisco, California, United States
Competitive base salary, founding-level equity, a performance bonus tied to customer deployments, and health, dental and vision coverage. Flexible arrangements for candidates outside the Bay Area are reviewed case by case.
HOW TO APPLYIf you want to own the infrastructure that frontier-level agent training runs on, we would like to hear from you. Apply through this platform, including your expected gross annual salary in USD and your availability to start.
