RL Systems Engineer for Frontier AI Tinker

Thinking Machines Lab Inc.; San Francisco (CA); On-site; USD 350,000 - 475,000; Full time

Benefits offered by this job

Health benefits

Unlimited PTO

Parental leave

Relocation support

Job summary

Thinking Machines Lab Inc. is hiring to build our training systems for Tinker, including RL systems, numerics, kernels, and beyond. You will work with researchers and external partners to push frontier models and enable post-training capabilities across the stack.

You’ll co-design RL algorithms, diagnose training runs, and optimize pipelines to deliver frontier-level results, making our platform and models perform at peak capability for researchers and developers alike.

Qualifications

• Bachelor’s degree or equivalent in CS/ML/Physics/Math with strong grounding.
• Proficiency in Python and DL frameworks; comfort debugging distributed training.
• Clarity in communication and ability to explain complex technical concepts.
• Strong interest in working on Tinker and increasing usefulness and adoption.
• Experience with RL training stability techniques preferred.
• Familiarity with low-precision training/inference and LLM serving stacks.

Responsibilities

• Develop frontier customization techniques and build the post-training engine.
• Engage with researchers and external partners pushing Tinker.
• Co-design RL algorithms and training systems across the stack.
• Debug RL runs, optimize post-training pipelines, and improve performance.

Skills

Python

DL frameworks

Distributed training

Communication

Interest in Tinker

Education

Bachelor's degree or equivalent

Tools

PyTorch

TensorFlow

JAX

SGLang

vLLM

Job description

Thinking Machines Lab Inc. is hiring to build our training systems for Tinker, including RL systems, numerics, kernels, and beyond. You will work with researchers and external partners to push frontier models and enable post-training capabilities across the stack.

You’ll co-design RL algorithms, diagnose training runs, and optimize pipelines to deliver frontier-level results, making our platform and models perform at peak capability for researchers and developers alike.