Remote Post-Training RL Engineer for LLMs

Promptcube3; New York (NY); Remote; USD 180,000 - 210,000; Full time

Benefits offered by this job

Weekly lunch stipend

Health and dental benefits

RRSP matching / 401K

Parental leave top-up

Enrichment benefits

Education stipend

Paid vacation (30 days)

Travel to other offices / offsite

Home office stipend

Job summary

Cohere is a leading security-first enterprise AI company building foundation models and end-to-end products. This remote role in Post-Training focuses on RL methods to train LLMs, improve generalization, and develop production-grade training recipes.

You’ll collaborate with post-training data infrastructure and evaluation teams to push the boundaries of model learning. The role supports remote work or office-based options across our global offices.

Qualifications

• Experience training large language models (LLMs).
• Proven ability to apply reinforcement learning to LLMs at scale.
• Strong experimental rigor and analysis with clear conclusions.

Responsibilities

• Design, run, and analyze large-scale RL post-training experiments (e.g., RLHF, RL from verifiable rewards).
• Research and develop methods to improve generalization and robustness.
• Build robust, reproducible training recipes for production models.
• Diagnose training dynamics and failure modes, including instability and reward over-optimization.
• Turn research into production-ready training pipelines.

Skills

LLM training

Reinforcement learning

Distributed training

Research & communication

Tools

PyTorch

JAX

Megatron

Job description

Cohere is a leading security-first enterprise AI company building foundation models and end-to-end products. This remote role in Post-Training focuses on RL methods to train LLMs, improve generalization, and develop production-grade training recipes.

You’ll collaborate with post-training data infrastructure and evaluation teams to push the boundaries of model learning. The role supports remote work or office-based options across our global offices.