Post-Training Data Infrastructure Engineer

Cohere; Oregon; Hybrid; USD 180,000 - 260,000; Full time

Benefits offered by this job

Lunch stipend for lunch

Health benefits

RRSP matching / 401K

Parental leave

Enrichment benefits

Education stipend

Vacation (30 days)

Travel to other offices

Home office stipend

Job summary

Cohere, a leading security-first enterprise AI company, is hiring a Member of Technical Staff for Post-Training Data Infrastructure. You will build end-to-end data pipelines for SFT, RL, and synthetic data, collaborating with researchers and annotation teams to accelerate model iteration.

You will own dataset quality, versioning, and reproducibility, and design schemas for efficient training of large language models. Remote-friendly with global offices and strong benefits.

Qualifications

• Strong software engineering skills.
• Experience designing and building large-scale data pipelines for machine learning, ideally for LLM post-training.
• Track record of running distributed data processing and model inference at scale.
• Excellent communication skills.

Responsibilities

• Build and maintain data pipelines for SFT, preference, and RL post-training data.
• Own dataset quality, versioning, and reproducibility.
• Turn research data recipes into production-grade, reusable components.
• Scale model inference for data generation and model-based grading.
• Design data formats and schemas that keep post-training datasets consistent and efficient to train on.
• Collaborate with researchers and engineers to ensure data pipelines keep pace with model needs.

Skills

Software engineering

Data pipelines

Distributed systems

LLM post-training

Communication

Tools

Python

SQL

Spark

Job description

Cohere, a leading security-first enterprise AI company, is hiring a Member of Technical Staff for Post-Training Data Infrastructure. You will build end-to-end data pipelines for SFT, RL, and synthetic data, collaborating with researchers and annotation teams to accelerate model iteration.

You will own dataset quality, versioning, and reproducibility, and design schemas for efficient training of large language models. Remote-friendly with global offices and strong benefits.