Software Engineer - Pretraining Data

Magic AI • San Francisco, CA, US • 1d 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 Software Engineer working on our pretraining data, you write efficient and robust pipelines for giant, multimodal datasets. You will develop and optimize web scraping techniques to harvest and maintain data at internet-scale.

What you might work on:

Design & implement multimodal (video, audio, text etc) web crawlers for scraping and indexing petabytes of data
Create large scale data processing pipelines using tools like Ray, Apache Spark, Apache Flink, Google BigQuery etc.
Implement and scale deduplication techniques across modalities and apply heuristic and model-based techniques for parsing and filtering crawled data
Identify new data sources for inclusion in pre/post-training datasets

What we’re looking for:

Strong proficiency in distributed computing and parallel processing techniques
Obsession with details, reliability, and good testing to ensure data quality and integrity
Experience with designing and maintaining high-performance, scalable data architectures
Ability to design, develop and operate an LLM data pipeline from web scraping to data loading

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.