RL Infrastructure Engineer — Scalable Training & Performance

xAI; Palo Alto (CA); On-site; USD 170,000 - 260,000; Full time

Benefits offered by this job

Health insurance

Life and AD&D insurance

Fertility benefits

Flexible vacation

Visa sponsorship

401(k) plan

Job summary

xAI in Palo Alto is seeking an experienced engineer to design and implement the RL training framework and the systems backing all RL workloads, from ablations to production runs. You will profile, debug, and optimize end-to-end training performance, and improve scalability and observability of the RL stack.

The role requires proficiency in Python and C++, familiarity with Jax or Rust, and experience with large-scale RL/NLP training infrastructures and RL numerics.

Qualifications

• Experience building, debugging, and optimizing efficiency of large-scale distributed systems.
• Comfortable diving into unfamiliar areas and solving problems at all levels of the stack.
• Proficiency in Python, Jax, Rust, and/or C++.
• Experience with large scale LLM training infrastructure.
• Strong knowledge of reinforcement learning techniques.
• Experience with RL numerics.

Responsibilities

• Design and implement the RL training framework and the systems backing RL workloads, from small-scale ablations to production training runs.
• Profile, debug, and optimize end-to-end training performance.
• Improve scalability and observability of the RL stack.
• Collaborate with the RL infrastructure team to ship reliable, scalable systems.

Skills

Python

Jax

Rust

C++

Reinforcement learning techniques

RL numerics

Job description

xAI in Palo Alto is seeking an experienced engineer to design and implement the RL training framework and the systems backing all RL workloads, from ablations to production runs. You will profile, debug, and optimize end-to-end training performance, and improve scalability and observability of the RL stack.

The role requires proficiency in Python and C++, familiarity with Jax or Rust, and experience with large-scale RL/NLP training infrastructures and RL numerics.