Post-Training RL Systems Engineer

Magic; San Francisco (CA); On-site; USD 200,000 - 550,000; Full time

Benefits offered by this job

Equity

401(k) with 6% match

Health insurance

Dental and vision insurance

Unlimited PTO

Visa sponsorship and relocation stipen

Small, focused team

Job summary

Magic is seeking a Software Engineer for the RL Research & Environments team to design and operate data, evaluation, and environment systems that improve model capabilities after pre-training.

You will focus on post-training: identifying capability gaps, building targeted datasets, designing reward signals, and running iterative training loops that measurably improve user-facing behavior. The role emphasizes infrastructure and experimental workflows for scalable RL.

Qualifications

• Strong software engineering fundamentals.
• Experience building or operating large-scale data or ML systems.
• Ability to design and interpret experiments that measure model behavior changes.
• Comfort working at the intersection of ML, data systems, and infrastructure.
• Strong attention to data quality and evaluation rigor.
• Track record of owning experimental or production systems end-to-end.

Responsibilities

• Design and operate data, evaluation, and environment systems to improve model capabilities after pre-training.
• Identify capability gaps and build targeted datasets, reward signals, and training loops.
• Own infrastructure and workflows that connect product priorities to capability gains.
• Build systems to surface long-context failure modes and enable rapid RL iteration at scale.
• Collaborate with Product and Research to translate capability goals into measurable iterations.

Skills

Software engineering fundamentals

Large-scale ML/data systems

Experiment design & interpretation

Data quality & evaluation rigor

End-to-end prod systems

Job description

Magic is seeking a Software Engineer for the RL Research & Environments team to design and operate data, evaluation, and environment systems that improve model capabilities after pre-training.

You will focus on post-training: identifying capability gaps, building targeted datasets, designing reward signals, and running iterative training loops that measurably improve user-facing behavior. The role emphasizes infrastructure and experimental workflows for scalable RL.