Applied Research Engineer - Post-training

San Francisco

Full Time Senior-level / Expert USD 100K - 550K

Magic

Magic is an AI company that is working toward building safe AGI to accelerate humanity’s progress on the world’s most important problems.

View all jobs at Magic

Apply now Apply later

Posted 2 weeks ago

Magic’s mission is to build safe AGI that accelerates humanity’s progress on the world’s most important problems. We believe the most promising path to safe AGI lies in automating research and code generation to improve models and solve alignment more reliably than humans can alone. Our approach combines frontier-scale pre-training, domain-specific RL, ultra-long context, and inference-time compute to achieve this goal.

About the role:

As a Research Engineer in post-training, you'll help develop novel techniques and datasets to maximize model performance for real-world applications, leveraging data and compute at scale. Working closely with the applied team and customers, you’ll enable our models to operate with the best possible effectiveness in large, real-world codebases.

What you might work on:

Research and develop innovative post-training techniques to improve models' ability to leverage additional compute and data at inference
Partner with applied teams to enhance model capabilities to generate substantial, high-quality, functional code while optimizing for an enjoyable user experience
Improve model capabilities for generating substantial, high-quality, functional code
Ensure seamless integration and intuitive workflows for our users
Create and refine continuous data feedback mechanisms to drive ongoing model improvements
Design scalable approaches to generate synthetic datasets

What we’re looking for:

Experience with the productization and deployment of a LLM based product
Strong general software engineering skills
Thorough knowledge of the deep learning literature
Experience with the post-training of LLMs
Ability to come up with and evaluate novel research ideas
Obsession with details, reliability, and good testing to ensure data quality and integrity
Willingness to dive deeply into a large ML codebase to debug

Magic strives to be the place where high-potential individuals can do their best work. We value quick learning and grit just as much as skill and experience.