Description
Magic is building frontier-scale code models around a combination of large-scale pre-training, domain-specific reinforcement learning, ultra-long context, and inference-time compute. The Research Engineer position sits across several of those layers rather than being limited to one narrow research problem.
The work can move from training trillion-parameter models on large GPU clusters to improving inference throughput for new architectures. Engineers may contribute to the research frameworks used for both experimentation and production, design or optimize model architectures, curate post-training data, and build internet-scale pipelines and crawlers that supply training workloads.
Large-model research at this scale also requires strong systems instincts. An experiment that looks promising algorithmically still has to fit within memory, networking, storage, and distributed-compute constraints. Candidates should therefore be comfortable working across deep-learning research and practical engineering rather than treating them as separate disciplines.
Magic is especially interested in people who understand modern deep-learning literature, can generate and test research ideas, and have practical experience with LLM pre-training or post-training. Experience operating large distributed systems and handling substantial ETL workloads is also highly relevant.
The position offers broad ownership inside a small research and engineering team. Someone joining at this stage may influence not only experiments but the infrastructure, datasets, evaluation practices, and model-development workflow used across future research.
Curated opportunity. Please verify details and apply via the original link below. No Signals are required for this project/job.
https://magic.dev/careers/bcc8d988-47ea-4089-9619-80260abb71b5