Member of Technical Staff, RL Research & Environments

About the role

Magic is developing safe AGI by focusing on automated research and code generation. You will join the RL Research & Environments team to build the infrastructure and data pipelines necessary for post-training, helping our models reason effectively over long contexts.

Key facts

What you'll do

• Create post-training datasets through synthetic generation, self-play, and manual collection.
• Develop and manage scoring, filtering, and mixture methods for RL training corpora.
• Construct evaluation systems that identify failure points in long-context model performance.
• Develop reward signals and training environments to drive specific model improvements.
• Execute ablation studies on data sources, reward structures, and task architectures.
• Enhance the observability and reliability of data and environment pipelines.
• Work with Research and Product teams to turn objectives into repeatable training cycles.

Requirements

• Proficiency in core software engineering principles.
• Background in operating or constructing large-scale machine learning or data systems.
• Ability to design experiments that quantify changes in model behavior.
• Experience working where infrastructure, data systems, and machine learning overlap.
• High standards for evaluation rigor and data quality.
• Proven ability to manage experimental or production systems from start to finish.

Skills & tools

• Reinforcement Learning
• Large-scale data systems
• Evaluation framework design
• Experimental design and analysis
• Infrastructure engineering

Practical notes

• Annual salary: $200,000 to $550,000 depending on experience.
• Significant equity grants included.
• Benefits: 401(k) with 6% matching, comprehensive health, dental, and vision insurance for employees and dependents, and unlimited paid time off.
• Visa sponsorship and relocation assistance are available for those moving to San Francisco.